egra: optimised SSE memfill and colorblend
[iv.d.git] / fixed16.d
blob3ad2aeccc988758e005e96340085eb1d68a86624
1 /* Invisible Vector Library
2 * coded by Ketmar // Invisible Vector <ketmar@ketmar.no-ip.org>
3 * Understanding is not required. Only obedience.
5 * This program is free software: you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, version 3 of the License ONLY.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program. If not, see <http://www.gnu.org/licenses/>.
17 // UNFINISHED! DO NOT USE!
18 module iv.fixed16;
20 // ////////////////////////////////////////////////////////////////////////// //
21 alias Fixnum = FixnumN!(int, 16);
22 enum Fixed(int v) = Fixnum(v);
23 enum Fixed(float v) = Fixnum(cast(double)v);
24 enum Fixed(double v) = Fixnum(v);
27 // ////////////////////////////////////////////////////////////////////////// //
28 struct FixnumN(T=int, ubyte scl=16) if ((is(T == byte) || is(T == int) || is(T == long)) && scl > 0 && scl < T.sizeof*8-1) {
29 private:
30 alias Me = typeof(this);
32 public:
33 enum mpl = cast(T)1<<scale;
34 enum scale = scl;
35 T value;
36 static FixnumN make (T v) nothrow @safe @nogc { FixnumN fixed; fixed.value = v; return fixed; }
38 public:
39 string toString () const { import std.format : format; return "%f".format(cast(double)value/mpl); }
41 public nothrow @safe @nogc:
42 // min and max represent the smallest and larget possible values respectively
43 enum min = make(-T.max);
44 enum max = make(T.max);
45 static if (is(T == int) && scl == 16) {
46 enum One = make(1<<16);
47 enum Half = make(1<<15);
48 enum Nan = make(0x80000000);
49 enum Inf = make(int.max);
50 enum PI = make(0x3243f);
51 enum PId2 = make(0x3243f>>1);
52 enum PIx2 = make(0x3243f<<1);
53 enum Sqrt2 = make(92682);
56 this (T v) { pragma(inline, true); value = v*mpl; }
57 this (double v) { pragma(inline, true); import core.math : rndtol; value = cast(T)rndtol(v*mpl); }
59 Me opUnary(string op:"++") () pure { pragma(inline, true); value += mpl; return this; }
60 Me opUnary(string op:"--") () pure { pragma(inline, true); value -= mpl; return this; }
62 Me opUnary(string op) () const pure { pragma(inline, true); mixin("return make("~op~"value);"); }
64 bool opEquals (Me b) const pure { pragma(inline, true); return (value == b.value); }
65 bool opEquals (T b) const pure { pragma(inline, true); return (value == b*mpl); }
66 bool opEquals (double b) const pure { pragma(inline, true); return ((cast(double)value/mpl) == b); }
68 int opCmp (const Me b) const pure { pragma(inline, true); return (value < b.value ? -1 : value > b.value ? 1 : 0); }
69 int opCmp (const T b) const pure { pragma(inline, true); return (value < b*mpl ? -1 : value > b*mpl ? 1 : 0); }
70 int opCmp (const double b) const pure { pragma(inline, true); return (value < b*mpl ? -1 : value > b*mpl ? 1 : 0); }
72 void opAssign (Me v) pure { pragma(inline, true); value = v.value; }
73 void opAssign (T v) pure { pragma(inline, true); value = v*mpl; }
74 void opAssign (double v) pure { pragma(inline, true); value = cast(T)(v*mpl); }
76 void opOpAssign (string op) (Me v) pure if (op == "+" || op == "-") { pragma(inline, true); mixin("value "~op~"= v.value;"); }
77 void opOpAssign (string op) (Me v) pure if (op == "*" || op == "/") { pragma(inline, true); mixin("value = cast(T)(cast(long)value"~op~"cast(long)v.value/mpl);"); }
79 void opOpAssign (string op) (T v) pure if (op == "+" || op == "-") { pragma(inline, true); mixin("value "~op~"= v*mpl;"); }
80 void opOpAssign (string op) (T v) pure if (op == "*" || op == "/") { pragma(inline, true); mixin("value "~op~"= v;"); }
82 void opOpAssign (string op) (double v) pure if (op == "+" || op == "-") { pragma(inline, true); import core.math : rndtol; mixin("value "~op~"= rndtol(v*mpl);"); }
83 void opOpAssign (string op) (double v) pure if (op == "*" || op == "/") { pragma(inline, true); import core.math : rndtol; mixin("value "~op~"= rndtol(v);"); }
85 T opCast(DT) () const pure if (is(DT == byte) || is(DT == ubyte) || is(DT == short) || is(DT == ushort) || is(DT == int) || is(DT == uint) || is(DT == long) || is(DT == ulong)) { pragma(inline, true); return cast(DT)(value/mpl); }
86 T opCast(DT) () const pure if (is(DT == float) || is(DT == double)) { pragma(inline, true); return cast(DT)((cast(double)value)/mpl); }
87 T opCast(DT) () const pure if (is(DT == bool)) { pragma(inline, true); return value != 0; }
90 Me opBinary(string op) (Me b) const pure if (op == "+" || op == "-") { pragma(inline, true); mixin("return make(value"~op~"b.value);"); }
91 Me opBinary(string op) (Me b) const pure if (op == "*" || op == "/") { pragma(inline, true); mixin("return make(cast(T)(cast(long)value"~op~"cast(long)b.value/mpl));"); }
93 Me opBinary(string op) (T b) const pure if (op == "+" || op == "-" || op == "%") { pragma(inline, true); mixin("return make(value"~op~"(b*mpl));"); }
94 Me opBinary(string op) (T b) const pure if (op == "*" || op == "/") { pragma(inline, true); mixin("return make(value"~op~"b);"); }
96 Me opBinaryRight(string op) (T b) const pure if (op == "+" || op == "-" || op == "%") { pragma(inline, true); mixin("return make((b*mpl)"~op~"value);"); }
97 Me opBinaryRight(string op) (T b) const pure if (op == "*") { pragma(inline, true); return make(cast(T)(b*value)); }
98 Me opBinaryRight(string op) (T b) const pure if (op == "/") { pragma(inline, true); return make(cast(T)(cast(long)(b*mpl*mpl)/value)); }
101 Me opBinary(string op) (double b) const pure if (op == "+" || op == "-" || op == "%") { pragma(inline, true); import core.math : rndtol; mixin("return make(value"~op~"cast(T)rndtol(b*mpl));"); }
102 Me opBinary(string op) (double b) const pure if (op == "*" || op == "/") { pragma(inline, true); import core.math : rndtol; mixin("return make(value"~op~"cast(T)rndtol(b));"); }
104 Me opBinaryRight(string op) (double b) const pure if (op == "+" || op == "-" || op == "%") { pragma(inline, true); import core.math : rndtol; mixin("return make(cast(T)rndtol(b*mpl)"~op~"value);"); }
105 Me opBinaryRight(string op) (double b) const pure if (op == "*") { pragma(inline, true); import core.math : rndtol; return make(cast(T)(cast(T)rndtol(b)*value)); }
106 Me opBinaryRight(string op) (double b) const pure if (op == "/") { pragma(inline, true); import core.math : rndtol; return make(cast(T)(rndtol(b*mpl*mpl)/value)); }
108 @property Me abs () const pure { pragma(inline, true); return make(value >= 0 ? value : -value); }
109 @property int sign () const pure { pragma(inline, true); return (value < 0 ? -1 : value > 0 ? 1 : 0); }
110 @property isnan () const pure { pragma(inline, true); return (value == T.min); }
112 static if (is(T == int) && scl == 16) {
113 static Me cos (Me rads) { pragma(inline, true); Me res; calcSinCosF16(rads.value, null, &res.value); return res; }
114 static Me sin (Me rads) { pragma(inline, true); Me res; calcSinCosF16(rads.value, &res.value, null); return res; }
115 static void sincos (Me rads, ref Me s, ref Me c) { pragma(inline, true); calcSinCosF16(rads.value, &s.value, &c.value); }
116 static Me atan2 (Me y, Me x) { pragma(inline, true); return make(calcAtan2F16(y.value, x.value)); }
117 static Me atan (Me x) { pragma(inline, true); return make(calcAtanF16(x.value)); }
118 static Me asin (Me x) { pragma(inline, true); return make(calcAsinF16(x.value)); }
121 private static:
122 // code taken from tbox
123 void calcSinCosF16 (int x, int* s, int* c) @trusted {
124 // |angle| < 90 degrees
125 // x0,y0: fixed30
126 static void cordicRotation (int* x0, int* y0, int z0) {
127 int i = 0;
128 int atan2i = 0;
129 int z = z0;
130 int x = *x0; // fixed30
131 int y = *y0; // fixed30
132 int xi = 0; // fixed30
133 int yi = 0; // fixed30
134 immutable(int)* patan2i = tb_fixed16_cordic_atan2i_table.ptr;
135 do {
136 xi = x>>i;
137 yi = y>>i;
138 atan2i = *patan2i++;
139 if (z >= 0) {
140 x -= yi;
141 y += xi;
142 z -= atan2i;
143 } else {
144 x += yi;
145 y -= xi;
146 z += atan2i;
148 } while (++i < 16);
149 *x0 = x;
150 *y0 = y;
153 // main
155 // (x0, y0) = (k, 0), k = 0.607252935 => fixed30
156 int cos = 0x26dd3b6a; // fixed30
157 int sin = 0; // fixed30
159 /* scale to 65536 degrees from x radians: x * 65536 / (2 * pi)
161 * 90: 0x40000000
162 * 180: 0x80000000
163 * 270: 0xc0000000
165 int ang = x*0x28be;
167 /* quadrant
169 * 1: 00 ...
170 * 2: 01 ...
171 * 3: 10 ...
172 * 4: 11 ...
174 * quadrant++
176 * 1: 01 ...
177 * 2: 10 ...
178 * 3: 11 ...
179 * 4: 00 ...
182 int quadrant = ang>>30;
183 ++quadrant;
185 /* quadrant == 2, 3, |angle| < 90
187 * 100 => -100 + 180 => 80
188 * -200 => 200 + 180 => -20
190 if (quadrant&0x2) ang = -ang+0x80000000;
192 // rotation
193 cordicRotation(&cos, &sin, ang);
195 // result
196 if (s !is null) *s = sin>>14; // fixed30->fixed16
197 if (c !is null) {
198 // quadrant == 2, 3
199 if (quadrant&0x2) cos = -cos;
200 *c = cos>>14; // fixed30->fixed16
204 // |angle| < 90 degrees
205 int cordicVectorAtan2 (int y0, int x0) @trusted {
206 int i = 0;
207 int atan2i = 0;
208 int z = 0;
209 int x = x0;
210 int y = y0;
211 int xi = 0;
212 int yi = 0;
213 immutable(int)* patan2i = tb_fixed16_cordic_atan2i_table.ptr;
214 do {
215 xi = x>>i;
216 yi = y>>i;
217 atan2i = *patan2i++;
218 if (y < 0) {
219 x -= yi;
220 y += xi;
221 z -= atan2i;
222 } else {
223 x += yi;
224 y -= xi;
225 z += atan2i;
227 } while (++i < 16);
228 return z / 0x28be;
231 // slope angle: [-180, 180]
232 // the precision will be pool if x, y is too small
233 int calcAtan2F16 (int y, int x) {
234 if (!(x|y)) return 0;
235 // abs
236 int xs = tbGetSign(x);
237 x = (x >= 0 ? x : -x); //tb_fixed30_abs(x);
238 // quadrant: 1, 4
239 int z = cordicVectorAtan2(y, x);
240 // for quadrant: 2, 3
241 if (xs) {
242 int zs = tbGetSign(z);
243 if (y == 0) zs = 0;
244 int pi = tbSetSign(0x3243f/*TB_FIXED16_PI*/, zs);
245 z = pi-z;
247 return z;
250 // |angle| < 90
251 // the precision will be pool if x is too large.
252 int calcAtanF16 (int x) {
253 if (!x) return 0;
254 return cordicVectorAtan2(x, 1<<16/*TB_FIXED16_ONE*/);
257 int calcAsinF16 (int x) {
258 // |angle| < 90 degrees
259 static int tb_fixed16_cordic_vector_asin (int m) @trusted {
260 int i = 0;
261 int atan2i = 0;
262 int z = 0;
263 int x = 0x18bde0bb; // k = 0.607252935
264 int y = 0;
265 int xi = 0;
266 int yi = 0;
267 immutable(int)* patan2i = tb_fixed16_cordic_atan2i_table.ptr;
268 do {
269 xi = x>>i;
270 yi = y>>i;
271 atan2i = *patan2i++;
272 if (y < m) {
273 x -= yi;
274 y += xi;
275 z -= atan2i;
276 } else {
277 x += yi;
278 y -= xi;
279 z += atan2i;
281 } while (++i < 16);
282 return z / 0x28be;
285 // abs
286 int s = tbGetSign(x);
287 x = (x >= 0 ? x : -1); //tb_fixed16_abs(x);
288 if (x >= 1<<16/*TB_FIXED16_ONE*/) return tbSetSign((0x3243f/*TB_FIXED16_PI*/) >> 1, s);
289 int z = tb_fixed16_cordic_vector_asin(x * 0x28be);
290 return tbSetSign(z, ~s);
293 // return -1 if x < 0, else return 0
294 int tbGetSign (int x) pure nothrow @safe @nogc {
295 pragma(inline, true);
296 int s = (cast(int)(x) >> 31);
297 //tb_assert((x < 0 && s == -1) || (x >= 0 && !s));
298 return s;
301 // if s == -1, return -x, else s must be 0, and return x.
302 int tbSetSign (int x, int s) pure nothrow @safe @nogc {
303 pragma(inline, true);
304 //tb_assert(s == 0 || s == -1);
305 return (x ^ s) - s;
309 int invert (int x) {
310 // is one?
311 if (x == 1<<16/*TB_FIXED16_ONE*/) return 1<<16/*TB_FIXED16_ONE*/;
312 // get sign
313 int s = tbGetSign(x);
314 // abs(x)
315 x = (x >= 0 ? x : -x);
316 // is infinity?
317 if (x <= 2) return (s < 0 ? -T.max : T.max); //return tbSetSign(TB_FIXED16_MAX, s);
319 // normalize
320 int cl0 = cast(int)tb_bits_cl0_u32_be(x);
321 x = x << cl0 >> 16;
323 // compute 1 / x approximation (0.5 <= x < 1.0)
324 // (2.90625 (~2.914) - 2 * x) >> 1
325 uint r = 0x17400-x;
327 // newton-raphson iteration:
328 // x = r * (2 - x * r) = ((r / 2) * (1 - x * r / 2)) * 4
329 r = ((0x10000 - ((x * r) >> 16)) * r) >> 15;
330 r = ((0x10000 - ((x * r) >> 16)) * r) >> (30 - cl0);
332 return tbSetSign(r, s);
336 private:
337 __gshared immutable int[30] tb_fixed16_cordic_atan2i_table = [
338 0x20000000 // 45.000000
339 , 0x12e4051d // 26.565051
340 , 0x9fb385b // 14.036243
341 , 0x51111d4 // 7.125016
342 , 0x28b0d43 // 3.576334
343 , 0x145d7e1 // 1.789911
344 , 0xa2f61e // 0.895174
345 , 0x517c55 // 0.447614
346 , 0x28be53 // 0.223811
347 , 0x145f2e // 0.111906
348 , 0xa2f98 // 0.055953
349 , 0x517cc // 0.027976
350 , 0x28be6 // 0.013988
351 , 0x145f3 // 0.006994
352 , 0xa2f9 // 0.003497
353 , 0x517c // 0.001749
354 , 0x28be // 0.000874
355 , 0x145f // 0.000437
356 , 0xa2f // 0.000219
357 , 0x517 // 0.000109
358 , 0x28b // 0.000055
359 , 0x145 // 0.000027
360 , 0xa2 // 0.000014
361 , 0x51 // 0.000007
362 , 0x28 // 0.000003
363 , 0x14 // 0.000002
364 , 0xa // 0.000001
365 , 0x5 // 0.000000
366 , 0x2 // 0.000000
367 , 0x1 // 0.000000
372 // ////////////////////////////////////////////////////////////////////////// //
374 import iv.vfs.io;
375 import std.math;
377 void main () {
378 writeln(Fixnum.PI);
379 //auto n = Fixed!42;
380 Fixnum n = 42;
381 Fixnum n1 = 42.2;
382 Fixnum nd1 = 0.1;
383 writeln(n);
384 writeln(n1);
385 n += 0.1;
386 writeln(n);
387 n += nd1;
388 writeln(n);
390 writeln("sin=", Fixnum.sin(Fixnum(0.3)));
391 writeln("dsin=", sin(0.3));
392 writeln("cos=", Fixnum.cos(Fixnum(0.5)));
393 writeln("dcos=", cos(0.5));