Source file src/math/big/float.go

     1  // Copyright 2014 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // This file implements multi-precision floating-point numbers.
     6  // Like in the GNU MPFR library (https://www.mpfr.org/), operands
     7  // can be of mixed precision. Unlike MPFR, the rounding mode is
     8  // not specified with each operation, but with each operand. The
     9  // rounding mode of the result operand determines the rounding
    10  // mode of an operation. This is a from-scratch implementation.
    11  
    12  package big
    13  
    14  import (
    15  	"fmt"
    16  	"math"
    17  	"math/bits"
    18  )
    19  
    20  const debugFloat = false // enable for debugging
    21  
    22  // A nonzero finite Float represents a multi-precision floating point number
    23  //
    24  //	sign × mantissa × 2**exponent
    25  //
    26  // with 0.5 <= mantissa < 1.0, and MinExp <= exponent <= MaxExp.
    27  // A Float may also be zero (+0, -0) or infinite (+Inf, -Inf).
    28  // All Floats are ordered, and the ordering of two Floats x and y
    29  // is defined by x.Cmp(y).
    30  //
    31  // Each Float value also has a precision, rounding mode, and accuracy.
    32  // The precision is the maximum number of mantissa bits available to
    33  // represent the value. The rounding mode specifies how a result should
    34  // be rounded to fit into the mantissa bits, and accuracy describes the
    35  // rounding error with respect to the exact result.
    36  //
    37  // Unless specified otherwise, all operations (including setters) that
    38  // specify a *Float variable for the result (usually via the receiver
    39  // with the exception of [Float.MantExp]), round the numeric result according
    40  // to the precision and rounding mode of the result variable.
    41  //
    42  // If the provided result precision is 0 (see below), it is set to the
    43  // precision of the argument with the largest precision value before any
    44  // rounding takes place, and the rounding mode remains unchanged. Thus,
    45  // uninitialized Floats provided as result arguments will have their
    46  // precision set to a reasonable value determined by the operands, and
    47  // their mode is the zero value for RoundingMode (ToNearestEven).
    48  //
    49  // By setting the desired precision to 24 or 53 and using matching rounding
    50  // mode (typically [ToNearestEven]), Float operations produce the same results
    51  // as the corresponding float32 or float64 IEEE 754 arithmetic for operands
    52  // that correspond to normal (i.e., not denormal) float32 or float64 numbers.
    53  // Exponent underflow and overflow lead to a 0 or an Infinity for different
    54  // values than IEEE 754 because Float exponents have a much larger range.
    55  //
    56  // The zero (uninitialized) value for a Float is ready to use and represents
    57  // the number +0.0 exactly, with precision 0 and rounding mode [ToNearestEven].
    58  //
    59  // Operations always take pointer arguments (*Float) rather
    60  // than Float values, and each unique Float value requires
    61  // its own unique *Float pointer. To "copy" a Float value,
    62  // an existing (or newly allocated) Float must be set to
    63  // a new value using the [Float.Set] method; shallow copies
    64  // of Floats are not supported and may lead to errors.
    65  type Float struct {
    66  	prec uint32
    67  	mode RoundingMode
    68  	acc  Accuracy
    69  	form form
    70  	neg  bool
    71  	mant nat
    72  	exp  int32
    73  }
    74  
    75  // An ErrNaN panic is raised by a [Float] operation that would lead to
    76  // a NaN under IEEE 754 rules. An ErrNaN implements the error interface.
    77  type ErrNaN struct {
    78  	msg string
    79  }
    80  
    81  func (err ErrNaN) Error() string {
    82  	return err.msg
    83  }
    84  
    85  // NewFloat allocates and returns a new [Float] set to x,
    86  // with precision 53 and rounding mode [ToNearestEven].
    87  // NewFloat panics with [ErrNaN] if x is a NaN.
    88  func NewFloat(x float64) *Float {
    89  	if math.IsNaN(x) {
    90  		panic(ErrNaN{"NewFloat(NaN)"})
    91  	}
    92  	return new(Float).SetFloat64(x)
    93  }
    94  
    95  // Exponent and precision limits.
    96  const (
    97  	MaxExp  = math.MaxInt32  // largest supported exponent
    98  	MinExp  = math.MinInt32  // smallest supported exponent
    99  	MaxPrec = math.MaxUint32 // largest (theoretically) supported precision; likely memory-limited
   100  )
   101  
   102  // Internal representation: The mantissa bits x.mant of a nonzero finite
   103  // Float x are stored in a nat slice long enough to hold up to x.prec bits;
   104  // the slice may (but doesn't have to) be shorter if the mantissa contains
   105  // trailing 0 bits. x.mant is normalized if the msb of x.mant == 1 (i.e.,
   106  // the msb is shifted all the way "to the left"). Thus, if the mantissa has
   107  // trailing 0 bits or x.prec is not a multiple of the Word size _W,
   108  // x.mant[0] has trailing zero bits. The msb of the mantissa corresponds
   109  // to the value 0.5; the exponent x.exp shifts the binary point as needed.
   110  //
   111  // A zero or non-finite Float x ignores x.mant and x.exp.
   112  //
   113  // x                 form      neg      mant         exp
   114  // ----------------------------------------------------------
   115  // ±0                zero      sign     -            -
   116  // 0 < |x| < +Inf    finite    sign     mantissa     exponent
   117  // ±Inf              inf       sign     -            -
   118  
   119  // A form value describes the internal representation.
   120  type form byte
   121  
   122  // The form value order is relevant - do not change!
   123  const (
   124  	zero form = iota
   125  	finite
   126  	inf
   127  )
   128  
   129  // RoundingMode determines how a [Float] value is rounded to the
   130  // desired precision. Rounding may change the [Float] value; the
   131  // rounding error is described by the [Float]'s [Accuracy].
   132  type RoundingMode byte
   133  
   134  // These constants define supported rounding modes.
   135  const (
   136  	ToNearestEven RoundingMode = iota // == IEEE 754-2008 roundTiesToEven
   137  	ToNearestAway                     // == IEEE 754-2008 roundTiesToAway
   138  	ToZero                            // == IEEE 754-2008 roundTowardZero
   139  	AwayFromZero                      // no IEEE 754-2008 equivalent
   140  	ToNegativeInf                     // == IEEE 754-2008 roundTowardNegative
   141  	ToPositiveInf                     // == IEEE 754-2008 roundTowardPositive
   142  )
   143  
   144  //go:generate stringer -type=RoundingMode
   145  
   146  // Accuracy describes the rounding error produced by the most recent
   147  // operation that generated a [Float] value, relative to the exact value.
   148  type Accuracy int8
   149  
   150  // Constants describing the [Accuracy] of a [Float].
   151  const (
   152  	Below Accuracy = -1
   153  	Exact Accuracy = 0
   154  	Above Accuracy = +1
   155  )
   156  
   157  //go:generate stringer -type=Accuracy
   158  
   159  // SetPrec sets z's precision to prec and returns the (possibly) rounded
   160  // value of z. Rounding occurs according to z's rounding mode if the mantissa
   161  // cannot be represented in prec bits without loss of precision.
   162  // SetPrec(0) maps all finite values to ±0; infinite values remain unchanged.
   163  // If prec > [MaxPrec], it is set to [MaxPrec].
   164  func (z *Float) SetPrec(prec uint) *Float {
   165  	z.acc = Exact // optimistically assume no rounding is needed
   166  
   167  	// special case
   168  	if prec == 0 {
   169  		z.prec = 0
   170  		if z.form == finite {
   171  			// truncate z to 0
   172  			z.acc = makeAcc(z.neg)
   173  			z.form = zero
   174  		}
   175  		return z
   176  	}
   177  
   178  	// general case
   179  	if prec > MaxPrec {
   180  		prec = MaxPrec
   181  	}
   182  	old := z.prec
   183  	z.prec = uint32(prec)
   184  	if z.prec < old {
   185  		z.round(0)
   186  	}
   187  	return z
   188  }
   189  
   190  func makeAcc(above bool) Accuracy {
   191  	if above {
   192  		return Above
   193  	}
   194  	return Below
   195  }
   196  
   197  // SetMode sets z's rounding mode to mode and returns an exact z.
   198  // z remains unchanged otherwise.
   199  // z.SetMode(z.Mode()) is a cheap way to set z's accuracy to [Exact].
   200  func (z *Float) SetMode(mode RoundingMode) *Float {
   201  	z.mode = mode
   202  	z.acc = Exact
   203  	return z
   204  }
   205  
   206  // Prec returns the mantissa precision of x in bits.
   207  // The result may be 0 for |x| == 0 and |x| == Inf.
   208  func (x *Float) Prec() uint {
   209  	return uint(x.prec)
   210  }
   211  
   212  // MinPrec returns the minimum precision required to represent x exactly
   213  // (i.e., the smallest prec before x.SetPrec(prec) would start rounding x).
   214  // The result is 0 for |x| == 0 and |x| == Inf.
   215  func (x *Float) MinPrec() uint {
   216  	if x.form != finite {
   217  		return 0
   218  	}
   219  	return uint(len(x.mant))*_W - x.mant.trailingZeroBits()
   220  }
   221  
   222  // Mode returns the rounding mode of x.
   223  func (x *Float) Mode() RoundingMode {
   224  	return x.mode
   225  }
   226  
   227  // Acc returns the accuracy of x produced by the most recent
   228  // operation, unless explicitly documented otherwise by that
   229  // operation.
   230  func (x *Float) Acc() Accuracy {
   231  	return x.acc
   232  }
   233  
   234  // Sign returns:
   235  //   - -1 if x < 0;
   236  //   - 0 if x is ±0;
   237  //   - +1 if x > 0.
   238  func (x *Float) Sign() int {
   239  	if debugFloat {
   240  		x.validate()
   241  	}
   242  	if x.form == zero {
   243  		return 0
   244  	}
   245  	if x.neg {
   246  		return -1
   247  	}
   248  	return 1
   249  }
   250  
   251  // MantExp breaks x into its mantissa and exponent components
   252  // and returns the exponent. If a non-nil mant argument is
   253  // provided its value is set to the mantissa of x, with the
   254  // same precision and rounding mode as x. The components
   255  // satisfy x == mant × 2**exp, with 0.5 <= |mant| < 1.0.
   256  // Calling MantExp with a nil argument is an efficient way to
   257  // get the exponent of the receiver.
   258  //
   259  // Special cases are:
   260  //
   261  //	(  ±0).MantExp(mant) = 0, with mant set to   ±0
   262  //	(±Inf).MantExp(mant) = 0, with mant set to ±Inf
   263  //
   264  // x and mant may be the same in which case x is set to its
   265  // mantissa value.
   266  func (x *Float) MantExp(mant *Float) (exp int) {
   267  	if debugFloat {
   268  		x.validate()
   269  	}
   270  	if x.form == finite {
   271  		exp = int(x.exp)
   272  	}
   273  	if mant != nil {
   274  		mant.Copy(x)
   275  		if mant.form == finite {
   276  			mant.exp = 0
   277  		}
   278  	}
   279  	return
   280  }
   281  
   282  func (z *Float) setExpAndRound(exp int64, sbit uint) {
   283  	if exp < MinExp {
   284  		// underflow
   285  		z.acc = makeAcc(z.neg)
   286  		z.form = zero
   287  		return
   288  	}
   289  
   290  	if exp > MaxExp {
   291  		// overflow
   292  		z.acc = makeAcc(!z.neg)
   293  		z.form = inf
   294  		return
   295  	}
   296  
   297  	z.form = finite
   298  	z.exp = int32(exp)
   299  	z.round(sbit)
   300  }
   301  
   302  // SetMantExp sets z to mant × 2**exp and returns z.
   303  // The result z has the same precision and rounding mode
   304  // as mant. SetMantExp is an inverse of [Float.MantExp] but does
   305  // not require 0.5 <= |mant| < 1.0. Specifically, for a
   306  // given x of type *[Float], SetMantExp relates to [Float.MantExp]
   307  // as follows:
   308  //
   309  //	mant := new(Float)
   310  //	new(Float).SetMantExp(mant, x.MantExp(mant)).Cmp(x) == 0
   311  //
   312  // Special cases are:
   313  //
   314  //	z.SetMantExp(  ±0, exp) =   ±0
   315  //	z.SetMantExp(±Inf, exp) = ±Inf
   316  //
   317  // z and mant may be the same in which case z's exponent
   318  // is set to exp.
   319  func (z *Float) SetMantExp(mant *Float, exp int) *Float {
   320  	if debugFloat {
   321  		z.validate()
   322  		mant.validate()
   323  	}
   324  	z.Copy(mant)
   325  
   326  	if z.form == finite {
   327  		// 0 < |mant| < +Inf
   328  		z.setExpAndRound(int64(z.exp)+int64(exp), 0)
   329  	}
   330  	return z
   331  }
   332  
   333  // Signbit reports whether x is negative or negative zero.
   334  func (x *Float) Signbit() bool {
   335  	return x.neg
   336  }
   337  
   338  // IsInf reports whether x is +Inf or -Inf.
   339  func (x *Float) IsInf() bool {
   340  	return x.form == inf
   341  }
   342  
   343  // IsInt reports whether x is an integer.
   344  // ±Inf values are not integers.
   345  func (x *Float) IsInt() bool {
   346  	if debugFloat {
   347  		x.validate()
   348  	}
   349  	// special cases
   350  	if x.form != finite {
   351  		return x.form == zero
   352  	}
   353  	// x.form == finite
   354  	if x.exp <= 0 {
   355  		return false
   356  	}
   357  	// x.exp > 0
   358  	return x.prec <= uint32(x.exp) || x.MinPrec() <= uint(x.exp) // not enough bits for fractional mantissa
   359  }
   360  
   361  // debugging support
   362  func (x *Float) validate() {
   363  	if !debugFloat {
   364  		// avoid performance bugs
   365  		panic("validate called but debugFloat is not set")
   366  	}
   367  	if msg := x.validate0(); msg != "" {
   368  		panic(msg)
   369  	}
   370  }
   371  
   372  func (x *Float) validate0() string {
   373  	if x.form != finite {
   374  		return ""
   375  	}
   376  	m := len(x.mant)
   377  	if m == 0 {
   378  		return "nonzero finite number with empty mantissa"
   379  	}
   380  	const msb = 1 << (_W - 1)
   381  	if x.mant[m-1]&msb == 0 {
   382  		return fmt.Sprintf("msb not set in last word %#x of %s", x.mant[m-1], x.Text('p', 0))
   383  	}
   384  	if x.prec == 0 {
   385  		return "zero precision finite number"
   386  	}
   387  	return ""
   388  }
   389  
   390  // round rounds z according to z.mode to z.prec bits and sets z.acc accordingly.
   391  // sbit must be 0 or 1 and summarizes any "sticky bit" information one might
   392  // have before calling round. z's mantissa must be normalized (with the msb set)
   393  // or empty.
   394  //
   395  // CAUTION: The rounding modes [ToNegativeInf], [ToPositiveInf] are affected by the
   396  // sign of z. For correct rounding, the sign of z must be set correctly before
   397  // calling round.
   398  func (z *Float) round(sbit uint) {
   399  	if debugFloat {
   400  		z.validate()
   401  	}
   402  
   403  	z.acc = Exact
   404  	if z.form != finite {
   405  		// ±0 or ±Inf => nothing left to do
   406  		return
   407  	}
   408  	// z.form == finite && len(z.mant) > 0
   409  	// m > 0 implies z.prec > 0 (checked by validate)
   410  
   411  	m := uint32(len(z.mant)) // present mantissa length in words
   412  	bits := m * _W           // present mantissa bits; bits > 0
   413  	if bits <= z.prec {
   414  		// mantissa fits => nothing to do
   415  		return
   416  	}
   417  	// bits > z.prec
   418  
   419  	// Rounding is based on two bits: the rounding bit (rbit) and the
   420  	// sticky bit (sbit). The rbit is the bit immediately before the
   421  	// z.prec leading mantissa bits (the "0.5"). The sbit is set if any
   422  	// of the bits before the rbit are set (the "0.25", "0.125", etc.):
   423  	//
   424  	//   rbit  sbit  => "fractional part"
   425  	//
   426  	//   0     0        == 0
   427  	//   0     1        >  0  , < 0.5
   428  	//   1     0        == 0.5
   429  	//   1     1        >  0.5, < 1.0
   430  
   431  	// bits > z.prec: mantissa too large => round
   432  	r := uint(bits - z.prec - 1) // rounding bit position; r >= 0
   433  	rbit := z.mant.bit(r) & 1    // rounding bit; be safe and ensure it's a single bit
   434  	// The sticky bit is only needed for rounding ToNearestEven
   435  	// or when the rounding bit is zero. Avoid computation otherwise.
   436  	if sbit == 0 && (rbit == 0 || z.mode == ToNearestEven) {
   437  		sbit = z.mant.sticky(r)
   438  	}
   439  	sbit &= 1 // be safe and ensure it's a single bit
   440  
   441  	// cut off extra words
   442  	n := (z.prec + (_W - 1)) / _W // mantissa length in words for desired precision
   443  	if m > n {
   444  		copy(z.mant, z.mant[m-n:]) // move n last words to front
   445  		z.mant = z.mant[:n]
   446  	}
   447  
   448  	// determine number of trailing zero bits (ntz) and compute lsb mask of mantissa's least-significant word
   449  	ntz := n*_W - z.prec // 0 <= ntz < _W
   450  	lsb := Word(1) << ntz
   451  
   452  	// round if result is inexact
   453  	if rbit|sbit != 0 {
   454  		// Make rounding decision: The result mantissa is truncated ("rounded down")
   455  		// by default. Decide if we need to increment, or "round up", the (unsigned)
   456  		// mantissa.
   457  		inc := false
   458  		switch z.mode {
   459  		case ToNegativeInf:
   460  			inc = z.neg
   461  		case ToZero:
   462  			// nothing to do
   463  		case ToNearestEven:
   464  			inc = rbit != 0 && (sbit != 0 || z.mant[0]&lsb != 0)
   465  		case ToNearestAway:
   466  			inc = rbit != 0
   467  		case AwayFromZero:
   468  			inc = true
   469  		case ToPositiveInf:
   470  			inc = !z.neg
   471  		default:
   472  			panic("unreachable")
   473  		}
   474  
   475  		// A positive result (!z.neg) is Above the exact result if we increment,
   476  		// and it's Below if we truncate (Exact results require no rounding).
   477  		// For a negative result (z.neg) it is exactly the opposite.
   478  		z.acc = makeAcc(inc != z.neg)
   479  
   480  		if inc {
   481  			// add 1 to mantissa
   482  			if addVW(z.mant, z.mant, lsb) != 0 {
   483  				// mantissa overflow => adjust exponent
   484  				if z.exp >= MaxExp {
   485  					// exponent overflow
   486  					z.form = inf
   487  					return
   488  				}
   489  				z.exp++
   490  				// adjust mantissa: divide by 2 to compensate for exponent adjustment
   491  				shrVU(z.mant, z.mant, 1)
   492  				// set msb == carry == 1 from the mantissa overflow above
   493  				const msb = 1 << (_W - 1)
   494  				z.mant[n-1] |= msb
   495  			}
   496  		}
   497  	}
   498  
   499  	// zero out trailing bits in least-significant word
   500  	z.mant[0] &^= lsb - 1
   501  
   502  	if debugFloat {
   503  		z.validate()
   504  	}
   505  }
   506  
   507  func (z *Float) setBits64(neg bool, x uint64) *Float {
   508  	if z.prec == 0 {
   509  		z.prec = 64
   510  	}
   511  	z.acc = Exact
   512  	z.neg = neg
   513  	if x == 0 {
   514  		z.form = zero
   515  		return z
   516  	}
   517  	// x != 0
   518  	z.form = finite
   519  	s := bits.LeadingZeros64(x)
   520  	z.mant = z.mant.setUint64(x << uint(s))
   521  	z.exp = int32(64 - s) // always fits
   522  	if z.prec < 64 {
   523  		z.round(0)
   524  	}
   525  	return z
   526  }
   527  
   528  // SetUint64 sets z to the (possibly rounded) value of x and returns z.
   529  // If z's precision is 0, it is changed to 64 (and rounding will have
   530  // no effect).
   531  func (z *Float) SetUint64(x uint64) *Float {
   532  	return z.setBits64(false, x)
   533  }
   534  
   535  // SetInt64 sets z to the (possibly rounded) value of x and returns z.
   536  // If z's precision is 0, it is changed to 64 (and rounding will have
   537  // no effect).
   538  func (z *Float) SetInt64(x int64) *Float {
   539  	u := x
   540  	if u < 0 {
   541  		u = -u
   542  	}
   543  	// We cannot simply call z.SetUint64(uint64(u)) and change
   544  	// the sign afterwards because the sign affects rounding.
   545  	return z.setBits64(x < 0, uint64(u))
   546  }
   547  
   548  // SetFloat64 sets z to the (possibly rounded) value of x and returns z.
   549  // If z's precision is 0, it is changed to 53 (and rounding will have
   550  // no effect). SetFloat64 panics with [ErrNaN] if x is a NaN.
   551  func (z *Float) SetFloat64(x float64) *Float {
   552  	if z.prec == 0 {
   553  		z.prec = 53
   554  	}
   555  	if math.IsNaN(x) {
   556  		panic(ErrNaN{"Float.SetFloat64(NaN)"})
   557  	}
   558  	z.acc = Exact
   559  	z.neg = math.Signbit(x) // handle -0, -Inf correctly
   560  	if x == 0 {
   561  		z.form = zero
   562  		return z
   563  	}
   564  	if math.IsInf(x, 0) {
   565  		z.form = inf
   566  		return z
   567  	}
   568  	// normalized x != 0
   569  	z.form = finite
   570  	fmant, exp := math.Frexp(x) // get normalized mantissa
   571  	z.mant = z.mant.setUint64(1<<63 | math.Float64bits(fmant)<<11)
   572  	z.exp = int32(exp) // always fits
   573  	if z.prec < 53 {
   574  		z.round(0)
   575  	}
   576  	return z
   577  }
   578  
   579  // fnorm normalizes mantissa m by shifting it to the left
   580  // such that the msb of the most-significant word (msw) is 1.
   581  // It returns the shift amount. It assumes that len(m) != 0.
   582  func fnorm(m nat) int64 {
   583  	if debugFloat && (len(m) == 0 || m[len(m)-1] == 0) {
   584  		panic("msw of mantissa is 0")
   585  	}
   586  	s := nlz(m[len(m)-1])
   587  	if s > 0 {
   588  		c := shlVU(m, m, s)
   589  		if debugFloat && c != 0 {
   590  			panic("nlz or shlVU incorrect")
   591  		}
   592  	}
   593  	return int64(s)
   594  }
   595  
   596  // SetInt sets z to the (possibly rounded) value of x and returns z.
   597  // If z's precision is 0, it is changed to the larger of x.BitLen()
   598  // or 64 (and rounding will have no effect).
   599  func (z *Float) SetInt(x *Int) *Float {
   600  	// TODO(gri) can be more efficient if z.prec > 0
   601  	// but small compared to the size of x, or if there
   602  	// are many trailing 0's.
   603  	bits := uint32(x.BitLen())
   604  	if z.prec == 0 {
   605  		z.prec = umax32(bits, 64)
   606  	}
   607  	z.acc = Exact
   608  	z.neg = x.neg
   609  	if len(x.abs) == 0 {
   610  		z.form = zero
   611  		return z
   612  	}
   613  	// x != 0
   614  	z.mant = z.mant.set(x.abs)
   615  	fnorm(z.mant)
   616  	z.setExpAndRound(int64(bits), 0)
   617  	return z
   618  }
   619  
   620  // SetRat sets z to the (possibly rounded) value of x and returns z.
   621  // If z's precision is 0, it is changed to the largest of a.BitLen(),
   622  // b.BitLen(), or 64; with x = a/b.
   623  func (z *Float) SetRat(x *Rat) *Float {
   624  	if x.IsInt() {
   625  		return z.SetInt(x.Num())
   626  	}
   627  	var a, b Float
   628  	a.SetInt(x.Num())
   629  	b.SetInt(x.Denom())
   630  	if z.prec == 0 {
   631  		z.prec = umax32(a.prec, b.prec)
   632  	}
   633  	return z.Quo(&a, &b)
   634  }
   635  
   636  // SetInf sets z to the infinite Float -Inf if signbit is
   637  // set, or +Inf if signbit is not set, and returns z. The
   638  // precision of z is unchanged and the result is always
   639  // [Exact].
   640  func (z *Float) SetInf(signbit bool) *Float {
   641  	z.acc = Exact
   642  	z.form = inf
   643  	z.neg = signbit
   644  	return z
   645  }
   646  
   647  // Set sets z to the (possibly rounded) value of x and returns z.
   648  // If z's precision is 0, it is changed to the precision of x
   649  // before setting z (and rounding will have no effect).
   650  // Rounding is performed according to z's precision and rounding
   651  // mode; and z's accuracy reports the result error relative to the
   652  // exact (not rounded) result.
   653  func (z *Float) Set(x *Float) *Float {
   654  	if debugFloat {
   655  		x.validate()
   656  	}
   657  	z.acc = Exact
   658  	if z != x {
   659  		z.form = x.form
   660  		z.neg = x.neg
   661  		if x.form == finite {
   662  			z.exp = x.exp
   663  			z.mant = z.mant.set(x.mant)
   664  		}
   665  		if z.prec == 0 {
   666  			z.prec = x.prec
   667  		} else if z.prec < x.prec {
   668  			z.round(0)
   669  		}
   670  	}
   671  	return z
   672  }
   673  
   674  // Copy sets z to x, with the same precision, rounding mode, and accuracy as x.
   675  // Copy returns z. If x and z are identical, Copy is a no-op.
   676  func (z *Float) Copy(x *Float) *Float {
   677  	if debugFloat {
   678  		x.validate()
   679  	}
   680  	if z != x {
   681  		z.prec = x.prec
   682  		z.mode = x.mode
   683  		z.acc = x.acc
   684  		z.form = x.form
   685  		z.neg = x.neg
   686  		if z.form == finite {
   687  			z.mant = z.mant.set(x.mant)
   688  			z.exp = x.exp
   689  		}
   690  	}
   691  	return z
   692  }
   693  
   694  // msb32 returns the 32 most significant bits of x.
   695  func msb32(x nat) uint32 {
   696  	i := len(x) - 1
   697  	if i < 0 {
   698  		return 0
   699  	}
   700  	if debugFloat && x[i]&(1<<(_W-1)) == 0 {
   701  		panic("x not normalized")
   702  	}
   703  	switch _W {
   704  	case 32:
   705  		return uint32(x[i])
   706  	case 64:
   707  		return uint32(x[i] >> 32)
   708  	}
   709  	panic("unreachable")
   710  }
   711  
   712  // msb64 returns the 64 most significant bits of x.
   713  func msb64(x nat) uint64 {
   714  	i := len(x) - 1
   715  	if i < 0 {
   716  		return 0
   717  	}
   718  	if debugFloat && x[i]&(1<<(_W-1)) == 0 {
   719  		panic("x not normalized")
   720  	}
   721  	switch _W {
   722  	case 32:
   723  		v := uint64(x[i]) << 32
   724  		if i > 0 {
   725  			v |= uint64(x[i-1])
   726  		}
   727  		return v
   728  	case 64:
   729  		return uint64(x[i])
   730  	}
   731  	panic("unreachable")
   732  }
   733  
   734  // Uint64 returns the unsigned integer resulting from truncating x
   735  // towards zero. If 0 <= x <= [math.MaxUint64], the result is [Exact]
   736  // if x is an integer and [Below] otherwise.
   737  // The result is (0, [Above]) for x < 0, and ([math.MaxUint64], [Below])
   738  // for x > [math.MaxUint64].
   739  func (x *Float) Uint64() (uint64, Accuracy) {
   740  	if debugFloat {
   741  		x.validate()
   742  	}
   743  
   744  	switch x.form {
   745  	case finite:
   746  		if x.neg {
   747  			return 0, Above
   748  		}
   749  		// 0 < x < +Inf
   750  		if x.exp <= 0 {
   751  			// 0 < x < 1
   752  			return 0, Below
   753  		}
   754  		// 1 <= x < Inf
   755  		if x.exp <= 64 {
   756  			// u = trunc(x) fits into a uint64
   757  			u := msb64(x.mant) >> (64 - uint32(x.exp))
   758  			if x.MinPrec() <= 64 {
   759  				return u, Exact
   760  			}
   761  			return u, Below // x truncated
   762  		}
   763  		// x too large
   764  		return math.MaxUint64, Below
   765  
   766  	case zero:
   767  		return 0, Exact
   768  
   769  	case inf:
   770  		if x.neg {
   771  			return 0, Above
   772  		}
   773  		return math.MaxUint64, Below
   774  	}
   775  
   776  	panic("unreachable")
   777  }
   778  
   779  // Int64 returns the integer resulting from truncating x towards zero.
   780  // If [math.MinInt64] <= x <= [math.MaxInt64], the result is [Exact] if x is
   781  // an integer, and [Above] (x < 0) or [Below] (x > 0) otherwise.
   782  // The result is ([math.MinInt64], [Above]) for x < [math.MinInt64],
   783  // and ([math.MaxInt64], [Below]) for x > [math.MaxInt64].
   784  func (x *Float) Int64() (int64, Accuracy) {
   785  	if debugFloat {
   786  		x.validate()
   787  	}
   788  
   789  	switch x.form {
   790  	case finite:
   791  		// 0 < |x| < +Inf
   792  		acc := makeAcc(x.neg)
   793  		if x.exp <= 0 {
   794  			// 0 < |x| < 1
   795  			return 0, acc
   796  		}
   797  		// x.exp > 0
   798  
   799  		// 1 <= |x| < +Inf
   800  		if x.exp <= 63 {
   801  			// i = trunc(x) fits into an int64 (excluding math.MinInt64)
   802  			i := int64(msb64(x.mant) >> (64 - uint32(x.exp)))
   803  			if x.neg {
   804  				i = -i
   805  			}
   806  			if x.MinPrec() <= uint(x.exp) {
   807  				return i, Exact
   808  			}
   809  			return i, acc // x truncated
   810  		}
   811  		if x.neg {
   812  			// check for special case x == math.MinInt64 (i.e., x == -(0.5 << 64))
   813  			if x.exp == 64 && x.MinPrec() == 1 {
   814  				acc = Exact
   815  			}
   816  			return math.MinInt64, acc
   817  		}
   818  		// x too large
   819  		return math.MaxInt64, Below
   820  
   821  	case zero:
   822  		return 0, Exact
   823  
   824  	case inf:
   825  		if x.neg {
   826  			return math.MinInt64, Above
   827  		}
   828  		return math.MaxInt64, Below
   829  	}
   830  
   831  	panic("unreachable")
   832  }
   833  
   834  // Float32 returns the float32 value nearest to x. If x is too small to be
   835  // represented by a float32 (|x| < [math.SmallestNonzeroFloat32]), the result
   836  // is (0, [Below]) or (-0, [Above]), respectively, depending on the sign of x.
   837  // If x is too large to be represented by a float32 (|x| > [math.MaxFloat32]),
   838  // the result is (+Inf, [Above]) or (-Inf, [Below]), depending on the sign of x.
   839  func (x *Float) Float32() (float32, Accuracy) {
   840  	if debugFloat {
   841  		x.validate()
   842  	}
   843  
   844  	switch x.form {
   845  	case finite:
   846  		// 0 < |x| < +Inf
   847  
   848  		const (
   849  			fbits = 32                //        float size
   850  			mbits = 23                //        mantissa size (excluding implicit msb)
   851  			ebits = fbits - mbits - 1 //     8  exponent size
   852  			bias  = 1<<(ebits-1) - 1  //   127  exponent bias
   853  			dmin  = 1 - bias - mbits  //  -149  smallest unbiased exponent (denormal)
   854  			emin  = 1 - bias          //  -126  smallest unbiased exponent (normal)
   855  			emax  = bias              //   127  largest unbiased exponent (normal)
   856  		)
   857  
   858  		// Float mantissa m is 0.5 <= m < 1.0; compute exponent e for float32 mantissa.
   859  		e := x.exp - 1 // exponent for normal mantissa m with 1.0 <= m < 2.0
   860  
   861  		// Compute precision p for float32 mantissa.
   862  		// If the exponent is too small, we have a denormal number before
   863  		// rounding and fewer than p mantissa bits of precision available
   864  		// (the exponent remains fixed but the mantissa gets shifted right).
   865  		p := mbits + 1 // precision of normal float
   866  		if e < emin {
   867  			// recompute precision
   868  			p = mbits + 1 - emin + int(e)
   869  			// If p == 0, the mantissa of x is shifted so much to the right
   870  			// that its msb falls immediately to the right of the float32
   871  			// mantissa space. In other words, if the smallest denormal is
   872  			// considered "1.0", for p == 0, the mantissa value m is >= 0.5.
   873  			// If m > 0.5, it is rounded up to 1.0; i.e., the smallest denormal.
   874  			// If m == 0.5, it is rounded down to even, i.e., 0.0.
   875  			// If p < 0, the mantissa value m is <= "0.25" which is never rounded up.
   876  			if p < 0 /* m <= 0.25 */ || p == 0 && x.mant.sticky(uint(len(x.mant))*_W-1) == 0 /* m == 0.5 */ {
   877  				// underflow to ±0
   878  				if x.neg {
   879  					var z float32
   880  					return -z, Above
   881  				}
   882  				return 0.0, Below
   883  			}
   884  			// otherwise, round up
   885  			// We handle p == 0 explicitly because it's easy and because
   886  			// Float.round doesn't support rounding to 0 bits of precision.
   887  			if p == 0 {
   888  				if x.neg {
   889  					return -math.SmallestNonzeroFloat32, Below
   890  				}
   891  				return math.SmallestNonzeroFloat32, Above
   892  			}
   893  		}
   894  		// p > 0
   895  
   896  		// round
   897  		var r Float
   898  		r.prec = uint32(p)
   899  		r.Set(x)
   900  		e = r.exp - 1
   901  
   902  		// Rounding may have caused r to overflow to ±Inf
   903  		// (rounding never causes underflows to 0).
   904  		// If the exponent is too large, also overflow to ±Inf.
   905  		if r.form == inf || e > emax {
   906  			// overflow
   907  			if x.neg {
   908  				return float32(math.Inf(-1)), Below
   909  			}
   910  			return float32(math.Inf(+1)), Above
   911  		}
   912  		// e <= emax
   913  
   914  		// Determine sign, biased exponent, and mantissa.
   915  		var sign, bexp, mant uint32
   916  		if x.neg {
   917  			sign = 1 << (fbits - 1)
   918  		}
   919  
   920  		// Rounding may have caused a denormal number to
   921  		// become normal. Check again.
   922  		if e < emin {
   923  			// denormal number: recompute precision
   924  			// Since rounding may have at best increased precision
   925  			// and we have eliminated p <= 0 early, we know p > 0.
   926  			// bexp == 0 for denormals
   927  			p = mbits + 1 - emin + int(e)
   928  			mant = msb32(r.mant) >> uint(fbits-p)
   929  		} else {
   930  			// normal number: emin <= e <= emax
   931  			bexp = uint32(e+bias) << mbits
   932  			mant = msb32(r.mant) >> ebits & (1<<mbits - 1) // cut off msb (implicit 1 bit)
   933  		}
   934  
   935  		return math.Float32frombits(sign | bexp | mant), r.acc
   936  
   937  	case zero:
   938  		if x.neg {
   939  			var z float32
   940  			return -z, Exact
   941  		}
   942  		return 0.0, Exact
   943  
   944  	case inf:
   945  		if x.neg {
   946  			return float32(math.Inf(-1)), Exact
   947  		}
   948  		return float32(math.Inf(+1)), Exact
   949  	}
   950  
   951  	panic("unreachable")
   952  }
   953  
   954  // Float64 returns the float64 value nearest to x. If x is too small to be
   955  // represented by a float64 (|x| < [math.SmallestNonzeroFloat64]), the result
   956  // is (0, [Below]) or (-0, [Above]), respectively, depending on the sign of x.
   957  // If x is too large to be represented by a float64 (|x| > [math.MaxFloat64]),
   958  // the result is (+Inf, [Above]) or (-Inf, [Below]), depending on the sign of x.
   959  func (x *Float) Float64() (float64, Accuracy) {
   960  	if debugFloat {
   961  		x.validate()
   962  	}
   963  
   964  	switch x.form {
   965  	case finite:
   966  		// 0 < |x| < +Inf
   967  
   968  		const (
   969  			fbits = 64                //        float size
   970  			mbits = 52                //        mantissa size (excluding implicit msb)
   971  			ebits = fbits - mbits - 1 //    11  exponent size
   972  			bias  = 1<<(ebits-1) - 1  //  1023  exponent bias
   973  			dmin  = 1 - bias - mbits  // -1074  smallest unbiased exponent (denormal)
   974  			emin  = 1 - bias          // -1022  smallest unbiased exponent (normal)
   975  			emax  = bias              //  1023  largest unbiased exponent (normal)
   976  		)
   977  
   978  		// Float mantissa m is 0.5 <= m < 1.0; compute exponent e for float64 mantissa.
   979  		e := x.exp - 1 // exponent for normal mantissa m with 1.0 <= m < 2.0
   980  
   981  		// Compute precision p for float64 mantissa.
   982  		// If the exponent is too small, we have a denormal number before
   983  		// rounding and fewer than p mantissa bits of precision available
   984  		// (the exponent remains fixed but the mantissa gets shifted right).
   985  		p := mbits + 1 // precision of normal float
   986  		if e < emin {
   987  			// recompute precision
   988  			p = mbits + 1 - emin + int(e)
   989  			// If p == 0, the mantissa of x is shifted so much to the right
   990  			// that its msb falls immediately to the right of the float64
   991  			// mantissa space. In other words, if the smallest denormal is
   992  			// considered "1.0", for p == 0, the mantissa value m is >= 0.5.
   993  			// If m > 0.5, it is rounded up to 1.0; i.e., the smallest denormal.
   994  			// If m == 0.5, it is rounded down to even, i.e., 0.0.
   995  			// If p < 0, the mantissa value m is <= "0.25" which is never rounded up.
   996  			if p < 0 /* m <= 0.25 */ || p == 0 && x.mant.sticky(uint(len(x.mant))*_W-1) == 0 /* m == 0.5 */ {
   997  				// underflow to ±0
   998  				if x.neg {
   999  					var z float64
  1000  					return -z, Above
  1001  				}
  1002  				return 0.0, Below
  1003  			}
  1004  			// otherwise, round up
  1005  			// We handle p == 0 explicitly because it's easy and because
  1006  			// Float.round doesn't support rounding to 0 bits of precision.
  1007  			if p == 0 {
  1008  				if x.neg {
  1009  					return -math.SmallestNonzeroFloat64, Below
  1010  				}
  1011  				return math.SmallestNonzeroFloat64, Above
  1012  			}
  1013  		}
  1014  		// p > 0
  1015  
  1016  		// round
  1017  		var r Float
  1018  		r.prec = uint32(p)
  1019  		r.Set(x)
  1020  		e = r.exp - 1
  1021  
  1022  		// Rounding may have caused r to overflow to ±Inf
  1023  		// (rounding never causes underflows to 0).
  1024  		// If the exponent is too large, also overflow to ±Inf.
  1025  		if r.form == inf || e > emax {
  1026  			// overflow
  1027  			if x.neg {
  1028  				return math.Inf(-1), Below
  1029  			}
  1030  			return math.Inf(+1), Above
  1031  		}
  1032  		// e <= emax
  1033  
  1034  		// Determine sign, biased exponent, and mantissa.
  1035  		var sign, bexp, mant uint64
  1036  		if x.neg {
  1037  			sign = 1 << (fbits - 1)
  1038  		}
  1039  
  1040  		// Rounding may have caused a denormal number to
  1041  		// become normal. Check again.
  1042  		if e < emin {
  1043  			// denormal number: recompute precision
  1044  			// Since rounding may have at best increased precision
  1045  			// and we have eliminated p <= 0 early, we know p > 0.
  1046  			// bexp == 0 for denormals
  1047  			p = mbits + 1 - emin + int(e)
  1048  			mant = msb64(r.mant) >> uint(fbits-p)
  1049  		} else {
  1050  			// normal number: emin <= e <= emax
  1051  			bexp = uint64(e+bias) << mbits
  1052  			mant = msb64(r.mant) >> ebits & (1<<mbits - 1) // cut off msb (implicit 1 bit)
  1053  		}
  1054  
  1055  		return math.Float64frombits(sign | bexp | mant), r.acc
  1056  
  1057  	case zero:
  1058  		if x.neg {
  1059  			var z float64
  1060  			return -z, Exact
  1061  		}
  1062  		return 0.0, Exact
  1063  
  1064  	case inf:
  1065  		if x.neg {
  1066  			return math.Inf(-1), Exact
  1067  		}
  1068  		return math.Inf(+1), Exact
  1069  	}
  1070  
  1071  	panic("unreachable")
  1072  }
  1073  
  1074  // Int returns the result of truncating x towards zero;
  1075  // or nil if x is an infinity.
  1076  // The result is [Exact] if x.IsInt(); otherwise it is [Below]
  1077  // for x > 0, and [Above] for x < 0.
  1078  // If a non-nil *[Int] argument z is provided, [Int] stores
  1079  // the result in z instead of allocating a new [Int].
  1080  func (x *Float) Int(z *Int) (*Int, Accuracy) {
  1081  	if debugFloat {
  1082  		x.validate()
  1083  	}
  1084  
  1085  	if z == nil && x.form <= finite {
  1086  		z = new(Int)
  1087  	}
  1088  
  1089  	switch x.form {
  1090  	case finite:
  1091  		// 0 < |x| < +Inf
  1092  		acc := makeAcc(x.neg)
  1093  		if x.exp <= 0 {
  1094  			// 0 < |x| < 1
  1095  			return z.SetInt64(0), acc
  1096  		}
  1097  		// x.exp > 0
  1098  
  1099  		// 1 <= |x| < +Inf
  1100  		// determine minimum required precision for x
  1101  		allBits := uint(len(x.mant)) * _W
  1102  		exp := uint(x.exp)
  1103  		if x.MinPrec() <= exp {
  1104  			acc = Exact
  1105  		}
  1106  		// shift mantissa as needed
  1107  		if z == nil {
  1108  			z = new(Int)
  1109  		}
  1110  		z.neg = x.neg
  1111  		switch {
  1112  		case exp > allBits:
  1113  			z.abs = z.abs.shl(x.mant, exp-allBits)
  1114  		default:
  1115  			z.abs = z.abs.set(x.mant)
  1116  		case exp < allBits:
  1117  			z.abs = z.abs.shr(x.mant, allBits-exp)
  1118  		}
  1119  		return z, acc
  1120  
  1121  	case zero:
  1122  		return z.SetInt64(0), Exact
  1123  
  1124  	case inf:
  1125  		return nil, makeAcc(x.neg)
  1126  	}
  1127  
  1128  	panic("unreachable")
  1129  }
  1130  
  1131  // Rat returns the rational number corresponding to x;
  1132  // or nil if x is an infinity.
  1133  // The result is [Exact] if x is not an Inf.
  1134  // If a non-nil *[Rat] argument z is provided, [Rat] stores
  1135  // the result in z instead of allocating a new [Rat].
  1136  func (x *Float) Rat(z *Rat) (*Rat, Accuracy) {
  1137  	if debugFloat {
  1138  		x.validate()
  1139  	}
  1140  
  1141  	if z == nil && x.form <= finite {
  1142  		z = new(Rat)
  1143  	}
  1144  
  1145  	switch x.form {
  1146  	case finite:
  1147  		// 0 < |x| < +Inf
  1148  		allBits := int32(len(x.mant)) * _W
  1149  		// build up numerator and denominator
  1150  		z.a.neg = x.neg
  1151  		switch {
  1152  		case x.exp > allBits:
  1153  			z.a.abs = z.a.abs.shl(x.mant, uint(x.exp-allBits))
  1154  			z.b.abs = z.b.abs[:0] // == 1 (see Rat)
  1155  			// z already in normal form
  1156  		default:
  1157  			z.a.abs = z.a.abs.set(x.mant)
  1158  			z.b.abs = z.b.abs[:0] // == 1 (see Rat)
  1159  			// z already in normal form
  1160  		case x.exp < allBits:
  1161  			z.a.abs = z.a.abs.set(x.mant)
  1162  			t := z.b.abs.setUint64(1)
  1163  			z.b.abs = t.shl(t, uint(allBits-x.exp))
  1164  			z.norm()
  1165  		}
  1166  		return z, Exact
  1167  
  1168  	case zero:
  1169  		return z.SetInt64(0), Exact
  1170  
  1171  	case inf:
  1172  		return nil, makeAcc(x.neg)
  1173  	}
  1174  
  1175  	panic("unreachable")
  1176  }
  1177  
  1178  // Abs sets z to the (possibly rounded) value |x| (the absolute value of x)
  1179  // and returns z.
  1180  func (z *Float) Abs(x *Float) *Float {
  1181  	z.Set(x)
  1182  	z.neg = false
  1183  	return z
  1184  }
  1185  
  1186  // Neg sets z to the (possibly rounded) value of x with its sign negated,
  1187  // and returns z.
  1188  func (z *Float) Neg(x *Float) *Float {
  1189  	z.Set(x)
  1190  	z.neg = !z.neg
  1191  	return z
  1192  }
  1193  
  1194  func validateBinaryOperands(x, y *Float) {
  1195  	if !debugFloat {
  1196  		// avoid performance bugs
  1197  		panic("validateBinaryOperands called but debugFloat is not set")
  1198  	}
  1199  	if len(x.mant) == 0 {
  1200  		panic("empty mantissa for x")
  1201  	}
  1202  	if len(y.mant) == 0 {
  1203  		panic("empty mantissa for y")
  1204  	}
  1205  }
  1206  
  1207  // z = x + y, ignoring signs of x and y for the addition
  1208  // but using the sign of z for rounding the result.
  1209  // x and y must have a non-empty mantissa and valid exponent.
  1210  func (z *Float) uadd(x, y *Float) {
  1211  	// Note: This implementation requires 2 shifts most of the
  1212  	// time. It is also inefficient if exponents or precisions
  1213  	// differ by wide margins. The following article describes
  1214  	// an efficient (but much more complicated) implementation
  1215  	// compatible with the internal representation used here:
  1216  	//
  1217  	// Vincent Lefèvre: "The Generic Multiple-Precision Floating-
  1218  	// Point Addition With Exact Rounding (as in the MPFR Library)"
  1219  	// http://www.vinc17.net/research/papers/rnc6.pdf
  1220  
  1221  	if debugFloat {
  1222  		validateBinaryOperands(x, y)
  1223  	}
  1224  
  1225  	// compute exponents ex, ey for mantissa with "binary point"
  1226  	// on the right (mantissa.0) - use int64 to avoid overflow
  1227  	ex := int64(x.exp) - int64(len(x.mant))*_W
  1228  	ey := int64(y.exp) - int64(len(y.mant))*_W
  1229  
  1230  	al := alias(z.mant, x.mant) || alias(z.mant, y.mant)
  1231  
  1232  	// TODO(gri) having a combined add-and-shift primitive
  1233  	//           could make this code significantly faster
  1234  	switch {
  1235  	case ex < ey:
  1236  		if al {
  1237  			t := nat(nil).shl(y.mant, uint(ey-ex))
  1238  			z.mant = z.mant.add(x.mant, t)
  1239  		} else {
  1240  			z.mant = z.mant.shl(y.mant, uint(ey-ex))
  1241  			z.mant = z.mant.add(x.mant, z.mant)
  1242  		}
  1243  	default:
  1244  		// ex == ey, no shift needed
  1245  		z.mant = z.mant.add(x.mant, y.mant)
  1246  	case ex > ey:
  1247  		if al {
  1248  			t := nat(nil).shl(x.mant, uint(ex-ey))
  1249  			z.mant = z.mant.add(t, y.mant)
  1250  		} else {
  1251  			z.mant = z.mant.shl(x.mant, uint(ex-ey))
  1252  			z.mant = z.mant.add(z.mant, y.mant)
  1253  		}
  1254  		ex = ey
  1255  	}
  1256  	// len(z.mant) > 0
  1257  
  1258  	z.setExpAndRound(ex+int64(len(z.mant))*_W-fnorm(z.mant), 0)
  1259  }
  1260  
  1261  // z = x - y for |x| > |y|, ignoring signs of x and y for the subtraction
  1262  // but using the sign of z for rounding the result.
  1263  // x and y must have a non-empty mantissa and valid exponent.
  1264  func (z *Float) usub(x, y *Float) {
  1265  	// This code is symmetric to uadd.
  1266  	// We have not factored the common code out because
  1267  	// eventually uadd (and usub) should be optimized
  1268  	// by special-casing, and the code will diverge.
  1269  
  1270  	if debugFloat {
  1271  		validateBinaryOperands(x, y)
  1272  	}
  1273  
  1274  	ex := int64(x.exp) - int64(len(x.mant))*_W
  1275  	ey := int64(y.exp) - int64(len(y.mant))*_W
  1276  
  1277  	al := alias(z.mant, x.mant) || alias(z.mant, y.mant)
  1278  
  1279  	switch {
  1280  	case ex < ey:
  1281  		if al {
  1282  			t := nat(nil).shl(y.mant, uint(ey-ex))
  1283  			z.mant = t.sub(x.mant, t)
  1284  		} else {
  1285  			z.mant = z.mant.shl(y.mant, uint(ey-ex))
  1286  			z.mant = z.mant.sub(x.mant, z.mant)
  1287  		}
  1288  	default:
  1289  		// ex == ey, no shift needed
  1290  		z.mant = z.mant.sub(x.mant, y.mant)
  1291  	case ex > ey:
  1292  		if al {
  1293  			t := nat(nil).shl(x.mant, uint(ex-ey))
  1294  			z.mant = t.sub(t, y.mant)
  1295  		} else {
  1296  			z.mant = z.mant.shl(x.mant, uint(ex-ey))
  1297  			z.mant = z.mant.sub(z.mant, y.mant)
  1298  		}
  1299  		ex = ey
  1300  	}
  1301  
  1302  	// operands may have canceled each other out
  1303  	if len(z.mant) == 0 {
  1304  		z.acc = Exact
  1305  		z.form = zero
  1306  		z.neg = false
  1307  		return
  1308  	}
  1309  	// len(z.mant) > 0
  1310  
  1311  	z.setExpAndRound(ex+int64(len(z.mant))*_W-fnorm(z.mant), 0)
  1312  }
  1313  
  1314  // z = x * y, ignoring signs of x and y for the multiplication
  1315  // but using the sign of z for rounding the result.
  1316  // x and y must have a non-empty mantissa and valid exponent.
  1317  func (z *Float) umul(x, y *Float) {
  1318  	if debugFloat {
  1319  		validateBinaryOperands(x, y)
  1320  	}
  1321  
  1322  	// Note: This is doing too much work if the precision
  1323  	// of z is less than the sum of the precisions of x
  1324  	// and y which is often the case (e.g., if all floats
  1325  	// have the same precision).
  1326  	// TODO(gri) Optimize this for the common case.
  1327  
  1328  	e := int64(x.exp) + int64(y.exp)
  1329  	if x == y {
  1330  		z.mant = z.mant.sqr(x.mant)
  1331  	} else {
  1332  		z.mant = z.mant.mul(x.mant, y.mant)
  1333  	}
  1334  	z.setExpAndRound(e-fnorm(z.mant), 0)
  1335  }
  1336  
  1337  // z = x / y, ignoring signs of x and y for the division
  1338  // but using the sign of z for rounding the result.
  1339  // x and y must have a non-empty mantissa and valid exponent.
  1340  func (z *Float) uquo(x, y *Float) {
  1341  	if debugFloat {
  1342  		validateBinaryOperands(x, y)
  1343  	}
  1344  
  1345  	// mantissa length in words for desired result precision + 1
  1346  	// (at least one extra bit so we get the rounding bit after
  1347  	// the division)
  1348  	n := int(z.prec/_W) + 1
  1349  
  1350  	// compute adjusted x.mant such that we get enough result precision
  1351  	xadj := x.mant
  1352  	if d := n - len(x.mant) + len(y.mant); d > 0 {
  1353  		// d extra words needed => add d "0 digits" to x
  1354  		xadj = make(nat, len(x.mant)+d)
  1355  		copy(xadj[d:], x.mant)
  1356  	}
  1357  	// TODO(gri): If we have too many digits (d < 0), we should be able
  1358  	// to shorten x for faster division. But we must be extra careful
  1359  	// with rounding in that case.
  1360  
  1361  	// Compute d before division since there may be aliasing of x.mant
  1362  	// (via xadj) or y.mant with z.mant.
  1363  	d := len(xadj) - len(y.mant)
  1364  
  1365  	// divide
  1366  	var r nat
  1367  	z.mant, r = z.mant.div(nil, xadj, y.mant)
  1368  	e := int64(x.exp) - int64(y.exp) - int64(d-len(z.mant))*_W
  1369  
  1370  	// The result is long enough to include (at least) the rounding bit.
  1371  	// If there's a non-zero remainder, the corresponding fractional part
  1372  	// (if it were computed), would have a non-zero sticky bit (if it were
  1373  	// zero, it couldn't have a non-zero remainder).
  1374  	var sbit uint
  1375  	if len(r) > 0 {
  1376  		sbit = 1
  1377  	}
  1378  
  1379  	z.setExpAndRound(e-fnorm(z.mant), sbit)
  1380  }
  1381  
  1382  // ucmp returns -1, 0, or +1, depending on whether
  1383  // |x| < |y|, |x| == |y|, or |x| > |y|.
  1384  // x and y must have a non-empty mantissa and valid exponent.
  1385  func (x *Float) ucmp(y *Float) int {
  1386  	if debugFloat {
  1387  		validateBinaryOperands(x, y)
  1388  	}
  1389  
  1390  	switch {
  1391  	case x.exp < y.exp:
  1392  		return -1
  1393  	case x.exp > y.exp:
  1394  		return +1
  1395  	}
  1396  	// x.exp == y.exp
  1397  
  1398  	// compare mantissas
  1399  	i := len(x.mant)
  1400  	j := len(y.mant)
  1401  	for i > 0 || j > 0 {
  1402  		var xm, ym Word
  1403  		if i > 0 {
  1404  			i--
  1405  			xm = x.mant[i]
  1406  		}
  1407  		if j > 0 {
  1408  			j--
  1409  			ym = y.mant[j]
  1410  		}
  1411  		switch {
  1412  		case xm < ym:
  1413  			return -1
  1414  		case xm > ym:
  1415  			return +1
  1416  		}
  1417  	}
  1418  
  1419  	return 0
  1420  }
  1421  
  1422  // Handling of sign bit as defined by IEEE 754-2008, section 6.3:
  1423  //
  1424  // When neither the inputs nor result are NaN, the sign of a product or
  1425  // quotient is the exclusive OR of the operands’ signs; the sign of a sum,
  1426  // or of a difference x−y regarded as a sum x+(−y), differs from at most
  1427  // one of the addends’ signs; and the sign of the result of conversions,
  1428  // the quantize operation, the roundToIntegral operations, and the
  1429  // roundToIntegralExact (see 5.3.1) is the sign of the first or only operand.
  1430  // These rules shall apply even when operands or results are zero or infinite.
  1431  //
  1432  // When the sum of two operands with opposite signs (or the difference of
  1433  // two operands with like signs) is exactly zero, the sign of that sum (or
  1434  // difference) shall be +0 in all rounding-direction attributes except
  1435  // roundTowardNegative; under that attribute, the sign of an exact zero
  1436  // sum (or difference) shall be −0. However, x+x = x−(−x) retains the same
  1437  // sign as x even when x is zero.
  1438  //
  1439  // See also: https://play.golang.org/p/RtH3UCt5IH
  1440  
  1441  // Add sets z to the rounded sum x+y and returns z. If z's precision is 0,
  1442  // it is changed to the larger of x's or y's precision before the operation.
  1443  // Rounding is performed according to z's precision and rounding mode; and
  1444  // z's accuracy reports the result error relative to the exact (not rounded)
  1445  // result. Add panics with [ErrNaN] if x and y are infinities with opposite
  1446  // signs. The value of z is undefined in that case.
  1447  func (z *Float) Add(x, y *Float) *Float {
  1448  	if debugFloat {
  1449  		x.validate()
  1450  		y.validate()
  1451  	}
  1452  
  1453  	if z.prec == 0 {
  1454  		z.prec = umax32(x.prec, y.prec)
  1455  	}
  1456  
  1457  	if x.form == finite && y.form == finite {
  1458  		// x + y (common case)
  1459  
  1460  		// Below we set z.neg = x.neg, and when z aliases y this will
  1461  		// change the y operand's sign. This is fine, because if an
  1462  		// operand aliases the receiver it'll be overwritten, but we still
  1463  		// want the original x.neg and y.neg values when we evaluate
  1464  		// x.neg != y.neg, so we need to save y.neg before setting z.neg.
  1465  		yneg := y.neg
  1466  
  1467  		z.neg = x.neg
  1468  		if x.neg == yneg {
  1469  			// x + y == x + y
  1470  			// (-x) + (-y) == -(x + y)
  1471  			z.uadd(x, y)
  1472  		} else {
  1473  			// x + (-y) == x - y == -(y - x)
  1474  			// (-x) + y == y - x == -(x - y)
  1475  			if x.ucmp(y) > 0 {
  1476  				z.usub(x, y)
  1477  			} else {
  1478  				z.neg = !z.neg
  1479  				z.usub(y, x)
  1480  			}
  1481  		}
  1482  		if z.form == zero && z.mode == ToNegativeInf && z.acc == Exact {
  1483  			z.neg = true
  1484  		}
  1485  		return z
  1486  	}
  1487  
  1488  	if x.form == inf && y.form == inf && x.neg != y.neg {
  1489  		// +Inf + -Inf
  1490  		// -Inf + +Inf
  1491  		// value of z is undefined but make sure it's valid
  1492  		z.acc = Exact
  1493  		z.form = zero
  1494  		z.neg = false
  1495  		panic(ErrNaN{"addition of infinities with opposite signs"})
  1496  	}
  1497  
  1498  	if x.form == zero && y.form == zero {
  1499  		// ±0 + ±0
  1500  		z.acc = Exact
  1501  		z.form = zero
  1502  		z.neg = x.neg && y.neg // -0 + -0 == -0
  1503  		return z
  1504  	}
  1505  
  1506  	if x.form == inf || y.form == zero {
  1507  		// ±Inf + y
  1508  		// x + ±0
  1509  		return z.Set(x)
  1510  	}
  1511  
  1512  	// ±0 + y
  1513  	// x + ±Inf
  1514  	return z.Set(y)
  1515  }
  1516  
  1517  // Sub sets z to the rounded difference x-y and returns z.
  1518  // Precision, rounding, and accuracy reporting are as for [Float.Add].
  1519  // Sub panics with [ErrNaN] if x and y are infinities with equal
  1520  // signs. The value of z is undefined in that case.
  1521  func (z *Float) Sub(x, y *Float) *Float {
  1522  	if debugFloat {
  1523  		x.validate()
  1524  		y.validate()
  1525  	}
  1526  
  1527  	if z.prec == 0 {
  1528  		z.prec = umax32(x.prec, y.prec)
  1529  	}
  1530  
  1531  	if x.form == finite && y.form == finite {
  1532  		// x - y (common case)
  1533  		yneg := y.neg
  1534  		z.neg = x.neg
  1535  		if x.neg != yneg {
  1536  			// x - (-y) == x + y
  1537  			// (-x) - y == -(x + y)
  1538  			z.uadd(x, y)
  1539  		} else {
  1540  			// x - y == x - y == -(y - x)
  1541  			// (-x) - (-y) == y - x == -(x - y)
  1542  			if x.ucmp(y) > 0 {
  1543  				z.usub(x, y)
  1544  			} else {
  1545  				z.neg = !z.neg
  1546  				z.usub(y, x)
  1547  			}
  1548  		}
  1549  		if z.form == zero && z.mode == ToNegativeInf && z.acc == Exact {
  1550  			z.neg = true
  1551  		}
  1552  		return z
  1553  	}
  1554  
  1555  	if x.form == inf && y.form == inf && x.neg == y.neg {
  1556  		// +Inf - +Inf
  1557  		// -Inf - -Inf
  1558  		// value of z is undefined but make sure it's valid
  1559  		z.acc = Exact
  1560  		z.form = zero
  1561  		z.neg = false
  1562  		panic(ErrNaN{"subtraction of infinities with equal signs"})
  1563  	}
  1564  
  1565  	if x.form == zero && y.form == zero {
  1566  		// ±0 - ±0
  1567  		z.acc = Exact
  1568  		z.form = zero
  1569  		z.neg = x.neg && !y.neg // -0 - +0 == -0
  1570  		return z
  1571  	}
  1572  
  1573  	if x.form == inf || y.form == zero {
  1574  		// ±Inf - y
  1575  		// x - ±0
  1576  		return z.Set(x)
  1577  	}
  1578  
  1579  	// ±0 - y
  1580  	// x - ±Inf
  1581  	return z.Neg(y)
  1582  }
  1583  
  1584  // Mul sets z to the rounded product x*y and returns z.
  1585  // Precision, rounding, and accuracy reporting are as for [Float.Add].
  1586  // Mul panics with [ErrNaN] if one operand is zero and the other
  1587  // operand an infinity. The value of z is undefined in that case.
  1588  func (z *Float) Mul(x, y *Float) *Float {
  1589  	if debugFloat {
  1590  		x.validate()
  1591  		y.validate()
  1592  	}
  1593  
  1594  	if z.prec == 0 {
  1595  		z.prec = umax32(x.prec, y.prec)
  1596  	}
  1597  
  1598  	z.neg = x.neg != y.neg
  1599  
  1600  	if x.form == finite && y.form == finite {
  1601  		// x * y (common case)
  1602  		z.umul(x, y)
  1603  		return z
  1604  	}
  1605  
  1606  	z.acc = Exact
  1607  	if x.form == zero && y.form == inf || x.form == inf && y.form == zero {
  1608  		// ±0 * ±Inf
  1609  		// ±Inf * ±0
  1610  		// value of z is undefined but make sure it's valid
  1611  		z.form = zero
  1612  		z.neg = false
  1613  		panic(ErrNaN{"multiplication of zero with infinity"})
  1614  	}
  1615  
  1616  	if x.form == inf || y.form == inf {
  1617  		// ±Inf * y
  1618  		// x * ±Inf
  1619  		z.form = inf
  1620  		return z
  1621  	}
  1622  
  1623  	// ±0 * y
  1624  	// x * ±0
  1625  	z.form = zero
  1626  	return z
  1627  }
  1628  
  1629  // Quo sets z to the rounded quotient x/y and returns z.
  1630  // Precision, rounding, and accuracy reporting are as for [Float.Add].
  1631  // Quo panics with [ErrNaN] if both operands are zero or infinities.
  1632  // The value of z is undefined in that case.
  1633  func (z *Float) Quo(x, y *Float) *Float {
  1634  	if debugFloat {
  1635  		x.validate()
  1636  		y.validate()
  1637  	}
  1638  
  1639  	if z.prec == 0 {
  1640  		z.prec = umax32(x.prec, y.prec)
  1641  	}
  1642  
  1643  	z.neg = x.neg != y.neg
  1644  
  1645  	if x.form == finite && y.form == finite {
  1646  		// x / y (common case)
  1647  		z.uquo(x, y)
  1648  		return z
  1649  	}
  1650  
  1651  	z.acc = Exact
  1652  	if x.form == zero && y.form == zero || x.form == inf && y.form == inf {
  1653  		// ±0 / ±0
  1654  		// ±Inf / ±Inf
  1655  		// value of z is undefined but make sure it's valid
  1656  		z.form = zero
  1657  		z.neg = false
  1658  		panic(ErrNaN{"division of zero by zero or infinity by infinity"})
  1659  	}
  1660  
  1661  	if x.form == zero || y.form == inf {
  1662  		// ±0 / y
  1663  		// x / ±Inf
  1664  		z.form = zero
  1665  		return z
  1666  	}
  1667  
  1668  	// x / ±0
  1669  	// ±Inf / y
  1670  	z.form = inf
  1671  	return z
  1672  }
  1673  
  1674  // Cmp compares x and y and returns:
  1675  //   - -1 if x < y;
  1676  //   - 0 if x == y (incl. -0 == 0, -Inf == -Inf, and +Inf == +Inf);
  1677  //   - +1 if x > y.
  1678  func (x *Float) Cmp(y *Float) int {
  1679  	if debugFloat {
  1680  		x.validate()
  1681  		y.validate()
  1682  	}
  1683  
  1684  	mx := x.ord()
  1685  	my := y.ord()
  1686  	switch {
  1687  	case mx < my:
  1688  		return -1
  1689  	case mx > my:
  1690  		return +1
  1691  	}
  1692  	// mx == my
  1693  
  1694  	// only if |mx| == 1 we have to compare the mantissae
  1695  	switch mx {
  1696  	case -1:
  1697  		return y.ucmp(x)
  1698  	case +1:
  1699  		return x.ucmp(y)
  1700  	}
  1701  
  1702  	return 0
  1703  }
  1704  
  1705  // ord classifies x and returns:
  1706  //
  1707  //	-2 if -Inf == x
  1708  //	-1 if -Inf < x < 0
  1709  //	 0 if x == 0 (signed or unsigned)
  1710  //	+1 if 0 < x < +Inf
  1711  //	+2 if x == +Inf
  1712  func (x *Float) ord() int {
  1713  	var m int
  1714  	switch x.form {
  1715  	case finite:
  1716  		m = 1
  1717  	case zero:
  1718  		return 0
  1719  	case inf:
  1720  		m = 2
  1721  	}
  1722  	if x.neg {
  1723  		m = -m
  1724  	}
  1725  	return m
  1726  }
  1727  
  1728  func umax32(x, y uint32) uint32 {
  1729  	if x > y {
  1730  		return x
  1731  	}
  1732  	return y
  1733  }
  1734  

View as plain text