...

Source file src/golang.org/x/image/vector/raster_fixed.go

Documentation: golang.org/x/image/vector

     1  // Copyright 2016 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  package vector
     6  
     7  // This file contains a fixed point math implementation of the vector
     8  // graphics rasterizer.
     9  
    10  const (
    11  	// ϕ is the number of binary digits after the fixed point.
    12  	//
    13  	// For example, if ϕ == 10 (and int1ϕ is based on the int32 type) then we
    14  	// are using 22.10 fixed point math.
    15  	//
    16  	// When changing this number, also change the assembly code (search for ϕ
    17  	// in the .s files).
    18  	ϕ = 9
    19  
    20  	fxOne          int1ϕ = 1 << ϕ
    21  	fxOneAndAHalf  int1ϕ = 1<<ϕ + 1<<(ϕ-1)
    22  	fxOneMinusIota int1ϕ = 1<<ϕ - 1 // Used for rounding up.
    23  )
    24  
    25  // int1ϕ is a signed fixed-point number with 1*ϕ binary digits after the fixed
    26  // point.
    27  type int1ϕ int32
    28  
    29  // int2ϕ is a signed fixed-point number with 2*ϕ binary digits after the fixed
    30  // point.
    31  //
    32  // The Rasterizer's bufU32 field, nominally of type []uint32 (since that slice
    33  // is also used by other code), can be thought of as a []int2ϕ during the
    34  // fixedLineTo method. Lines of code that are actually like:
    35  //
    36  //	buf[i] += uint32(etc) // buf has type []uint32.
    37  //
    38  // can be thought of as
    39  //
    40  //	buf[i] += int2ϕ(etc)  // buf has type []int2ϕ.
    41  type int2ϕ int32
    42  
    43  func fixedMax(x, y int1ϕ) int1ϕ {
    44  	if x > y {
    45  		return x
    46  	}
    47  	return y
    48  }
    49  
    50  func fixedMin(x, y int1ϕ) int1ϕ {
    51  	if x < y {
    52  		return x
    53  	}
    54  	return y
    55  }
    56  
    57  func fixedFloor(x int1ϕ) int32 { return int32(x >> ϕ) }
    58  func fixedCeil(x int1ϕ) int32  { return int32((x + fxOneMinusIota) >> ϕ) }
    59  
    60  func (z *Rasterizer) fixedLineTo(bx, by float32) {
    61  	ax, ay := z.penX, z.penY
    62  	z.penX, z.penY = bx, by
    63  	dir := int1ϕ(1)
    64  	if ay > by {
    65  		dir, ax, ay, bx, by = -1, bx, by, ax, ay
    66  	}
    67  	// Horizontal line segments yield no change in coverage. Almost horizontal
    68  	// segments would yield some change, in ideal math, but the computation
    69  	// further below, involving 1 / (by - ay), is unstable in fixed point math,
    70  	// so we treat the segment as if it was perfectly horizontal.
    71  	if by-ay <= 0.000001 {
    72  		return
    73  	}
    74  	dxdy := (bx - ax) / (by - ay)
    75  
    76  	ayϕ := int1ϕ(ay * float32(fxOne))
    77  	byϕ := int1ϕ(by * float32(fxOne))
    78  
    79  	x := int1ϕ(ax * float32(fxOne))
    80  	y := fixedFloor(ayϕ)
    81  	yMax := fixedCeil(byϕ)
    82  	if yMax > int32(z.size.Y) {
    83  		yMax = int32(z.size.Y)
    84  	}
    85  	width := int32(z.size.X)
    86  
    87  	for ; y < yMax; y++ {
    88  		dy := fixedMin(int1ϕ(y+1)<<ϕ, byϕ) - fixedMax(int1ϕ(y)<<ϕ, ayϕ)
    89  		xNext := x + int1ϕ(float32(dy)*dxdy)
    90  		if y < 0 {
    91  			x = xNext
    92  			continue
    93  		}
    94  		buf := z.bufU32[y*width:]
    95  		d := dy * dir // d ranges up to ±1<<(1*ϕ).
    96  		x0, x1 := x, xNext
    97  		if x > xNext {
    98  			x0, x1 = x1, x0
    99  		}
   100  		x0i := fixedFloor(x0)
   101  		x0Floor := int1ϕ(x0i) << ϕ
   102  		x1i := fixedCeil(x1)
   103  		x1Ceil := int1ϕ(x1i) << ϕ
   104  
   105  		if x1i <= x0i+1 {
   106  			xmf := (x+xNext)>>1 - x0Floor
   107  			if i := clamp(x0i+0, width); i < uint(len(buf)) {
   108  				buf[i] += uint32(d * (fxOne - xmf))
   109  			}
   110  			if i := clamp(x0i+1, width); i < uint(len(buf)) {
   111  				buf[i] += uint32(d * xmf)
   112  			}
   113  		} else {
   114  			oneOverS := x1 - x0
   115  			twoOverS := 2 * oneOverS
   116  			x0f := x0 - x0Floor
   117  			oneMinusX0f := fxOne - x0f
   118  			oneMinusX0fSquared := oneMinusX0f * oneMinusX0f
   119  			x1f := x1 - x1Ceil + fxOne
   120  			x1fSquared := x1f * x1f
   121  
   122  			// These next two variables are unused, as rounding errors are
   123  			// minimized when we delay the division by oneOverS for as long as
   124  			// possible. These lines of code (and the "In ideal math" comments
   125  			// below) are commented out instead of deleted in order to aid the
   126  			// comparison with the floating point version of the rasterizer.
   127  			//
   128  			// a0 := ((oneMinusX0f * oneMinusX0f) >> 1) / oneOverS
   129  			// am := ((x1f * x1f) >> 1) / oneOverS
   130  
   131  			if i := clamp(x0i, width); i < uint(len(buf)) {
   132  				// In ideal math: buf[i] += uint32(d * a0)
   133  				D := oneMinusX0fSquared // D ranges up to ±1<<(2*ϕ).
   134  				D *= d                  // D ranges up to ±1<<(3*ϕ).
   135  				D /= twoOverS
   136  				buf[i] += uint32(D)
   137  			}
   138  
   139  			if x1i == x0i+2 {
   140  				if i := clamp(x0i+1, width); i < uint(len(buf)) {
   141  					// In ideal math: buf[i] += uint32(d * (fxOne - a0 - am))
   142  					//
   143  					// (x1i == x0i+2) and (twoOverS == 2 * (x1 - x0)) implies
   144  					// that twoOverS ranges up to +1<<(1*ϕ+2).
   145  					D := twoOverS<<ϕ - oneMinusX0fSquared - x1fSquared // D ranges up to ±1<<(2*ϕ+2).
   146  					D *= d                                             // D ranges up to ±1<<(3*ϕ+2).
   147  					D /= twoOverS
   148  					buf[i] += uint32(D)
   149  				}
   150  			} else {
   151  				// This is commented out for the same reason as a0 and am.
   152  				//
   153  				// a1 := ((fxOneAndAHalf - x0f) << ϕ) / oneOverS
   154  
   155  				if i := clamp(x0i+1, width); i < uint(len(buf)) {
   156  					// In ideal math:
   157  					//	buf[i] += uint32(d * (a1 - a0))
   158  					// or equivalently (but better in non-ideal, integer math,
   159  					// with respect to rounding errors),
   160  					//	buf[i] += uint32(A * d / twoOverS)
   161  					// where
   162  					//	A = (a1 - a0) * twoOverS
   163  					//	  = a1*twoOverS - a0*twoOverS
   164  					// Noting that twoOverS/oneOverS equals 2, substituting for
   165  					// a0 and then a1, given above, yields:
   166  					//	A = a1*twoOverS - oneMinusX0fSquared
   167  					//	  = (fxOneAndAHalf-x0f)<<(ϕ+1) - oneMinusX0fSquared
   168  					//	  = fxOneAndAHalf<<(ϕ+1) - x0f<<(ϕ+1) - oneMinusX0fSquared
   169  					//
   170  					// This is a positive number minus two non-negative
   171  					// numbers. For an upper bound on A, the positive number is
   172  					//	P = fxOneAndAHalf<<(ϕ+1)
   173  					//	  < (2*fxOne)<<(ϕ+1)
   174  					//	  = fxOne<<(ϕ+2)
   175  					//	  = 1<<(2*ϕ+2)
   176  					//
   177  					// For a lower bound on A, the two non-negative numbers are
   178  					//	N = x0f<<(ϕ+1) + oneMinusX0fSquared
   179  					//	  ≤ x0f<<(ϕ+1) + fxOne*fxOne
   180  					//	  = x0f<<(ϕ+1) + 1<<(2*ϕ)
   181  					//	  < x0f<<(ϕ+1) + 1<<(2*ϕ+1)
   182  					//	  ≤ fxOne<<(ϕ+1) + 1<<(2*ϕ+1)
   183  					//	  = 1<<(2*ϕ+1) + 1<<(2*ϕ+1)
   184  					//	  = 1<<(2*ϕ+2)
   185  					//
   186  					// Thus, A ranges up to ±1<<(2*ϕ+2). It is possible to
   187  					// derive a tighter bound, but this bound is sufficient to
   188  					// reason about overflow.
   189  					D := (fxOneAndAHalf-x0f)<<(ϕ+1) - oneMinusX0fSquared // D ranges up to ±1<<(2*ϕ+2).
   190  					D *= d                                               // D ranges up to ±1<<(3*ϕ+2).
   191  					D /= twoOverS
   192  					buf[i] += uint32(D)
   193  				}
   194  				dTimesS := uint32((d << (2 * ϕ)) / oneOverS)
   195  				for xi := x0i + 2; xi < x1i-1; xi++ {
   196  					if i := clamp(xi, width); i < uint(len(buf)) {
   197  						buf[i] += dTimesS
   198  					}
   199  				}
   200  
   201  				// This is commented out for the same reason as a0 and am.
   202  				//
   203  				// a2 := a1 + (int1ϕ(x1i-x0i-3)<<(2*ϕ))/oneOverS
   204  
   205  				if i := clamp(x1i-1, width); i < uint(len(buf)) {
   206  					// In ideal math:
   207  					//	buf[i] += uint32(d * (fxOne - a2 - am))
   208  					// or equivalently (but better in non-ideal, integer math,
   209  					// with respect to rounding errors),
   210  					//	buf[i] += uint32(A * d / twoOverS)
   211  					// where
   212  					//	A = (fxOne - a2 - am) * twoOverS
   213  					//	  = twoOverS<<ϕ - a2*twoOverS - am*twoOverS
   214  					// Noting that twoOverS/oneOverS equals 2, substituting for
   215  					// am and then a2, given above, yields:
   216  					//	A = twoOverS<<ϕ - a2*twoOverS - x1f*x1f
   217  					//	  = twoOverS<<ϕ - a1*twoOverS - (int1ϕ(x1i-x0i-3)<<(2*ϕ))*2 - x1f*x1f
   218  					//	  = twoOverS<<ϕ - a1*twoOverS - int1ϕ(x1i-x0i-3)<<(2*ϕ+1) - x1f*x1f
   219  					// Substituting for a1, given above, yields:
   220  					//	A = twoOverS<<ϕ - ((fxOneAndAHalf-x0f)<<ϕ)*2 - int1ϕ(x1i-x0i-3)<<(2*ϕ+1) - x1f*x1f
   221  					//	  = twoOverS<<ϕ - (fxOneAndAHalf-x0f)<<(ϕ+1) - int1ϕ(x1i-x0i-3)<<(2*ϕ+1) - x1f*x1f
   222  					//	  = B<<ϕ - x1f*x1f
   223  					// where
   224  					//	B = twoOverS - (fxOneAndAHalf-x0f)<<1 - int1ϕ(x1i-x0i-3)<<(ϕ+1)
   225  					//	  = (x1-x0)<<1 - (fxOneAndAHalf-x0f)<<1 - int1ϕ(x1i-x0i-3)<<(ϕ+1)
   226  					//
   227  					// Re-arranging the defintions given above:
   228  					//	x0Floor := int1ϕ(x0i) << ϕ
   229  					//	x0f := x0 - x0Floor
   230  					//	x1Ceil := int1ϕ(x1i) << ϕ
   231  					//	x1f := x1 - x1Ceil + fxOne
   232  					// combined with fxOne = 1<<ϕ yields:
   233  					//	x0 = x0f + int1ϕ(x0i)<<ϕ
   234  					//	x1 = x1f + int1ϕ(x1i-1)<<ϕ
   235  					// so that expanding (x1-x0) yields:
   236  					//	B = (x1f-x0f + int1ϕ(x1i-x0i-1)<<ϕ)<<1 - (fxOneAndAHalf-x0f)<<1 - int1ϕ(x1i-x0i-3)<<(ϕ+1)
   237  					//	  = (x1f-x0f)<<1 + int1ϕ(x1i-x0i-1)<<(ϕ+1) - (fxOneAndAHalf-x0f)<<1 - int1ϕ(x1i-x0i-3)<<(ϕ+1)
   238  					// A large part of the second and fourth terms cancel:
   239  					//	B = (x1f-x0f)<<1 - (fxOneAndAHalf-x0f)<<1 - int1ϕ(-2)<<(ϕ+1)
   240  					//	  = (x1f-x0f)<<1 - (fxOneAndAHalf-x0f)<<1 + 1<<(ϕ+2)
   241  					//	  = (x1f - fxOneAndAHalf)<<1 + 1<<(ϕ+2)
   242  					// The first term, (x1f - fxOneAndAHalf)<<1, is a negative
   243  					// number, bounded below by -fxOneAndAHalf<<1, which is
   244  					// greater than -fxOne<<2, or -1<<(ϕ+2). Thus, B ranges up
   245  					// to ±1<<(ϕ+2). One final simplification:
   246  					//	B = x1f<<1 + (1<<(ϕ+2) - fxOneAndAHalf<<1)
   247  					const C = 1<<(ϕ+2) - fxOneAndAHalf<<1
   248  					D := x1f<<1 + C // D ranges up to ±1<<(1*ϕ+2).
   249  					D <<= ϕ         // D ranges up to ±1<<(2*ϕ+2).
   250  					D -= x1fSquared // D ranges up to ±1<<(2*ϕ+3).
   251  					D *= d          // D ranges up to ±1<<(3*ϕ+3).
   252  					D /= twoOverS
   253  					buf[i] += uint32(D)
   254  				}
   255  			}
   256  
   257  			if i := clamp(x1i, width); i < uint(len(buf)) {
   258  				// In ideal math: buf[i] += uint32(d * am)
   259  				D := x1fSquared // D ranges up to ±1<<(2*ϕ).
   260  				D *= d          // D ranges up to ±1<<(3*ϕ).
   261  				D /= twoOverS
   262  				buf[i] += uint32(D)
   263  			}
   264  		}
   265  
   266  		x = xNext
   267  	}
   268  }
   269  
   270  func fixedAccumulateOpOver(dst []uint8, src []uint32) {
   271  	// Sanity check that len(dst) >= len(src).
   272  	if len(dst) < len(src) {
   273  		return
   274  	}
   275  
   276  	acc := int2ϕ(0)
   277  	for i, v := range src {
   278  		acc += int2ϕ(v)
   279  		a := acc
   280  		if a < 0 {
   281  			a = -a
   282  		}
   283  		a >>= 2*ϕ - 16
   284  		if a > 0xffff {
   285  			a = 0xffff
   286  		}
   287  		// This algorithm comes from the standard library's image/draw package.
   288  		dstA := uint32(dst[i]) * 0x101
   289  		maskA := uint32(a)
   290  		outA := dstA*(0xffff-maskA)/0xffff + maskA
   291  		dst[i] = uint8(outA >> 8)
   292  	}
   293  }
   294  
   295  func fixedAccumulateOpSrc(dst []uint8, src []uint32) {
   296  	// Sanity check that len(dst) >= len(src).
   297  	if len(dst) < len(src) {
   298  		return
   299  	}
   300  
   301  	acc := int2ϕ(0)
   302  	for i, v := range src {
   303  		acc += int2ϕ(v)
   304  		a := acc
   305  		if a < 0 {
   306  			a = -a
   307  		}
   308  		a >>= 2*ϕ - 8
   309  		if a > 0xff {
   310  			a = 0xff
   311  		}
   312  		dst[i] = uint8(a)
   313  	}
   314  }
   315  
   316  func fixedAccumulateMask(buf []uint32) {
   317  	acc := int2ϕ(0)
   318  	for i, v := range buf {
   319  		acc += int2ϕ(v)
   320  		a := acc
   321  		if a < 0 {
   322  			a = -a
   323  		}
   324  		a >>= 2*ϕ - 16
   325  		if a > 0xffff {
   326  			a = 0xffff
   327  		}
   328  		buf[i] = uint32(a)
   329  	}
   330  }
   331  

View as plain text