Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(58)

Side by Side Diff: third_party/fdlibm/fdlibm.js

Issue 457643002: Implement Math.log1p using port from fdlibm. (Closed) Base URL: https://v8.googlecode.com/svn/branches/bleeding_edge
Patch Set: changed test Created 6 years, 4 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch | Annotate | Revision Log
« no previous file with comments | « third_party/fdlibm/fdlibm.cc ('k') | tools/generate-runtime-tests.py » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 // The following is adapted from fdlibm (http://www.netlib.org/fdlibm), 1 // The following is adapted from fdlibm (http://www.netlib.org/fdlibm),
2 // 2 //
3 // ==================================================== 3 // ====================================================
4 // Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. 4 // Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
5 // 5 //
6 // Developed at SunSoft, a Sun Microsystems, Inc. business. 6 // Developed at SunSoft, a Sun Microsystems, Inc. business.
7 // Permission to use, copy, modify, and distribute this 7 // Permission to use, copy, modify, and distribute this
8 // software is freely granted, provided that this notice 8 // software is freely granted, provided that this notice
9 // is preserved. 9 // is preserved.
10 // ==================================================== 10 // ====================================================
11 // 11 //
12 // The original source code covered by the above license above has been 12 // The original source code covered by the above license above has been
13 // modified significantly by Google Inc. 13 // modified significantly by Google Inc.
14 // Copyright 2014 the V8 project authors. All rights reserved. 14 // Copyright 2014 the V8 project authors. All rights reserved.
15 // 15 //
16 // The following is a straightforward translation of fdlibm routines for 16 // The following is a straightforward translation of fdlibm routines
17 // sin, cos, and tan, by Raymond Toy (rtoy@google.com). 17 // by Raymond Toy (rtoy@google.com).
18 18
19 19
20 var kTrig; // Initialized to a Float64Array during genesis and is not writable. 20 var kMath; // Initialized to a Float64Array during genesis and is not writable.
21 21
22 const INVPIO2 = kTrig[0]; 22 const INVPIO2 = kMath[0];
23 const PIO2_1 = kTrig[1]; 23 const PIO2_1 = kMath[1];
24 const PIO2_1T = kTrig[2]; 24 const PIO2_1T = kMath[2];
25 const PIO2_2 = kTrig[3]; 25 const PIO2_2 = kMath[3];
26 const PIO2_2T = kTrig[4]; 26 const PIO2_2T = kMath[4];
27 const PIO2_3 = kTrig[5]; 27 const PIO2_3 = kMath[5];
28 const PIO2_3T = kTrig[6]; 28 const PIO2_3T = kMath[6];
29 const PIO4 = kTrig[32]; 29 const PIO4 = kMath[32];
30 const PIO4LO = kTrig[33]; 30 const PIO4LO = kMath[33];
31 31
32 // Compute k and r such that x - k*pi/2 = r where |r| < pi/4. For 32 // Compute k and r such that x - k*pi/2 = r where |r| < pi/4. For
33 // precision, r is returned as two values y0 and y1 such that r = y0 + y1 33 // precision, r is returned as two values y0 and y1 such that r = y0 + y1
34 // to more than double precision. 34 // to more than double precision.
35 macro REMPIO2(X) 35 macro REMPIO2(X)
36 var n, y0, y1; 36 var n, y0, y1;
37 var hx = %_DoubleHi(X); 37 var hx = %_DoubleHi(X);
38 var ix = hx & 0x7fffffff; 38 var ix = hx & 0x7fffffff;
39 39
40 if (ix < 0x4002d97c) { 40 if (ix < 0x4002d97c) {
(...skipping 85 matching lines...) Expand 10 before | Expand all | Expand 10 after
126 // 126 //
127 // 3. ieee_sin(X+Y) = ieee_sin(X) + sin'(X')*Y 127 // 3. ieee_sin(X+Y) = ieee_sin(X) + sin'(X')*Y
128 // ~ ieee_sin(X) + (1-X*X/2)*Y 128 // ~ ieee_sin(X) + (1-X*X/2)*Y
129 // For better accuracy, let 129 // For better accuracy, let
130 // 3 2 2 2 2 130 // 3 2 2 2 2
131 // r = X *(S2+X *(S3+X *(S4+X *(S5+X *S6)))) 131 // r = X *(S2+X *(S3+X *(S4+X *(S5+X *S6))))
132 // then 3 2 132 // then 3 2
133 // sin(x) = X + (S1*X + (X *(r-Y/2)+Y)) 133 // sin(x) = X + (S1*X + (X *(r-Y/2)+Y))
134 // 134 //
135 macro KSIN(x) 135 macro KSIN(x)
136 kTrig[7+x] 136 kMath[7+x]
137 endmacro 137 endmacro
138 138
139 macro RETURN_KERNELSIN(X, Y, SIGN) 139 macro RETURN_KERNELSIN(X, Y, SIGN)
140 var z = X * X; 140 var z = X * X;
141 var v = z * X; 141 var v = z * X;
142 var r = KSIN(1) + z * (KSIN(2) + z * (KSIN(3) + 142 var r = KSIN(1) + z * (KSIN(2) + z * (KSIN(3) +
143 z * (KSIN(4) + z * KSIN(5)))); 143 z * (KSIN(4) + z * KSIN(5))));
144 return (X - ((z * (0.5 * Y - v * r) - Y) - v * KSIN(0))) SIGN; 144 return (X - ((z * (0.5 * Y - v * r) - Y) - v * KSIN(0))) SIGN;
145 endmacro 145 endmacro
146 146
(...skipping 23 matching lines...) Expand all
170 // cos(X+Y) = 1 - (X*X/2 - (r - X*Y)) 170 // cos(X+Y) = 1 - (X*X/2 - (r - X*Y))
171 // For better accuracy when x > 0.3, let qx = |x|/4 with 171 // For better accuracy when x > 0.3, let qx = |x|/4 with
172 // the last 32 bits mask off, and if x > 0.78125, let qx = 0.28125. 172 // the last 32 bits mask off, and if x > 0.78125, let qx = 0.28125.
173 // Then 173 // Then
174 // cos(X+Y) = (1-qx) - ((X*X/2-qx) - (r-X*Y)). 174 // cos(X+Y) = (1-qx) - ((X*X/2-qx) - (r-X*Y)).
175 // Note that 1-qx and (X*X/2-qx) is EXACT here, and the 175 // Note that 1-qx and (X*X/2-qx) is EXACT here, and the
176 // magnitude of the latter is at least a quarter of X*X/2, 176 // magnitude of the latter is at least a quarter of X*X/2,
177 // thus, reducing the rounding error in the subtraction. 177 // thus, reducing the rounding error in the subtraction.
178 // 178 //
179 macro KCOS(x) 179 macro KCOS(x)
180 kTrig[13+x] 180 kMath[13+x]
181 endmacro 181 endmacro
182 182
183 macro RETURN_KERNELCOS(X, Y, SIGN) 183 macro RETURN_KERNELCOS(X, Y, SIGN)
184 var ix = %_DoubleHi(X) & 0x7fffffff; 184 var ix = %_DoubleHi(X) & 0x7fffffff;
185 var z = X * X; 185 var z = X * X;
186 var r = z * (KCOS(0) + z * (KCOS(1) + z * (KCOS(2)+ 186 var r = z * (KCOS(0) + z * (KCOS(1) + z * (KCOS(2)+
187 z * (KCOS(3) + z * (KCOS(4) + z * KCOS(5)))))); 187 z * (KCOS(3) + z * (KCOS(4) + z * KCOS(5))))));
188 if (ix < 0x3fd33333) { // |x| ~< 0.3 188 if (ix < 0x3fd33333) { // |x| ~< 0.3
189 return (1 - (0.5 * z - (z * r - X * Y))) SIGN; 189 return (1 - (0.5 * z - (z * r - X * Y))) SIGN;
190 } else { 190 } else {
191 var qx; 191 var qx;
192 if (ix > 0x3fe90000) { // |x| > 0.78125 192 if (ix > 0x3fe90000) { // |x| > 0.78125
193 qx = 0.28125; 193 qx = 0.28125;
194 } else { 194 } else {
195 qx = %_ConstructDouble(%_DoubleHi(0.25 * X), 0); 195 qx = %_ConstructDouble(%_DoubleHi(0.25 * X), 0);
196 } 196 }
197 var hz = 0.5 * z - qx; 197 var hz = 0.5 * z - qx;
198 return (1 - qx - (hz - (z * r - X * Y))) SIGN; 198 return (1 - qx - (hz - (z * r - X * Y))) SIGN;
199 } 199 }
200 endmacro 200 endmacro
201 201
202
202 // kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854 203 // kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854
203 // Input x is assumed to be bounded by ~pi/4 in magnitude. 204 // Input x is assumed to be bounded by ~pi/4 in magnitude.
204 // Input y is the tail of x. 205 // Input y is the tail of x.
205 // Input k indicates whether ieee_tan (if k = 1) or -1/tan (if k = -1) 206 // Input k indicates whether ieee_tan (if k = 1) or -1/tan (if k = -1)
206 // is returned. 207 // is returned.
207 // 208 //
208 // Algorithm 209 // Algorithm
209 // 1. Since ieee_tan(-x) = -ieee_tan(x), we need only to consider positive x. 210 // 1. Since ieee_tan(-x) = -ieee_tan(x), we need only to consider positive x.
210 // 2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0. 211 // 2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0.
211 // 3. ieee_tan(x) is approximated by a odd polynomial of degree 27 on 212 // 3. ieee_tan(x) is approximated by a odd polynomial of degree 27 on
(...skipping 16 matching lines...) Expand all
228 // tan(x+y) = x + (T1*x + (x *(r+y)+y)) 229 // tan(x+y) = x + (T1*x + (x *(r+y)+y))
229 // 230 //
230 // 4. For x in [0.67434,pi/4], let y = pi/4 - x, then 231 // 4. For x in [0.67434,pi/4], let y = pi/4 - x, then
231 // tan(x) = ieee_tan(pi/4-y) = (1-ieee_tan(y))/(1+ieee_tan(y)) 232 // tan(x) = ieee_tan(pi/4-y) = (1-ieee_tan(y))/(1+ieee_tan(y))
232 // = 1 - 2*(ieee_tan(y) - (ieee_tan(y)^2)/(1+ieee_tan(y))) 233 // = 1 - 2*(ieee_tan(y) - (ieee_tan(y)^2)/(1+ieee_tan(y)))
233 // 234 //
234 // Set returnTan to 1 for tan; -1 for cot. Anything else is illegal 235 // Set returnTan to 1 for tan; -1 for cot. Anything else is illegal
235 // and will cause incorrect results. 236 // and will cause incorrect results.
236 // 237 //
237 macro KTAN(x) 238 macro KTAN(x)
238 kTrig[19+x] 239 kMath[19+x]
239 endmacro 240 endmacro
240 241
241 function KernelTan(x, y, returnTan) { 242 function KernelTan(x, y, returnTan) {
242 var z; 243 var z;
243 var w; 244 var w;
244 var hx = %_DoubleHi(x); 245 var hx = %_DoubleHi(x);
245 var ix = hx & 0x7fffffff; 246 var ix = hx & 0x7fffffff;
246 247
247 if (ix < 0x3e300000) { // |x| < 2^-28 248 if (ix < 0x3e300000) { // |x| < 2^-28
248 if (((ix | %_DoubleLo(x)) | (returnTan + 1)) == 0) { 249 if (((ix | %_DoubleLo(x)) | (returnTan + 1)) == 0) {
(...skipping 98 matching lines...) Expand 10 before | Expand all | Expand 10 after
347 // ECMA 262 - 15.8.2.18 348 // ECMA 262 - 15.8.2.18
348 function MathTan(x) { 349 function MathTan(x) {
349 x = x * 1; // Convert to number. 350 x = x * 1; // Convert to number.
350 if ((%_DoubleHi(x) & 0x7fffffff) <= 0x3fe921fb) { 351 if ((%_DoubleHi(x) & 0x7fffffff) <= 0x3fe921fb) {
351 // |x| < pi/4, approximately. No reduction needed. 352 // |x| < pi/4, approximately. No reduction needed.
352 return KernelTan(x, 0, 1); 353 return KernelTan(x, 0, 1);
353 } 354 }
354 REMPIO2(x); 355 REMPIO2(x);
355 return KernelTan(y0, y1, (n & 1) ? -1 : 1); 356 return KernelTan(y0, y1, (n & 1) ? -1 : 1);
356 } 357 }
358
359 // ES6 draft 09-27-13, section 20.2.2.20.
360 // Math.log1p
361 //
362 // Method :
363 // 1. Argument Reduction: find k and f such that
364 // 1+x = 2^k * (1+f),
365 // where sqrt(2)/2 < 1+f < sqrt(2) .
366 //
367 // Note. If k=0, then f=x is exact. However, if k!=0, then f
368 // may not be representable exactly. In that case, a correction
369 // term is need. Let u=1+x rounded. Let c = (1+x)-u, then
370 // log(1+x) - log(u) ~ c/u. Thus, we proceed to compute log(u),
371 // and add back the correction term c/u.
372 // (Note: when x > 2**53, one can simply return log(x))
373 //
374 // 2. Approximation of log1p(f).
375 // Let s = f/(2+f) ; based on log(1+f) = log(1+s) - log(1-s)
376 // = 2s + 2/3 s**3 + 2/5 s**5 + .....,
377 // = 2s + s*R
378 // We use a special Reme algorithm on [0,0.1716] to generate
379 // a polynomial of degree 14 to approximate R The maximum error
380 // of this polynomial approximation is bounded by 2**-58.45. In
381 // other words,
382 // 2 4 6 8 10 12 14
383 // R(z) ~ Lp1*s +Lp2*s +Lp3*s +Lp4*s +Lp5*s +Lp6*s +Lp7*s
384 // (the values of Lp1 to Lp7 are listed in the program)
385 // and
386 // | 2 14 | -58.45
387 // | Lp1*s +...+Lp7*s - R(z) | <= 2
388 // | |
389 // Note that 2s = f - s*f = f - hfsq + s*hfsq, where hfsq = f*f/2.
390 // In order to guarantee error in log below 1ulp, we compute log
391 // by
392 // log1p(f) = f - (hfsq - s*(hfsq+R)).
393 //
394 // 3. Finally, log1p(x) = k*ln2 + log1p(f).
395 // = k*ln2_hi+(f-(hfsq-(s*(hfsq+R)+k*ln2_lo)))
396 // Here ln2 is split into two floating point number:
397 // ln2_hi + ln2_lo,
398 // where n*ln2_hi is always exact for |n| < 2000.
399 //
400 // Special cases:
401 // log1p(x) is NaN with signal if x < -1 (including -INF) ;
402 // log1p(+INF) is +INF; log1p(-1) is -INF with signal;
403 // log1p(NaN) is that NaN with no signal.
404 //
405 // Accuracy:
406 // according to an error analysis, the error is always less than
407 // 1 ulp (unit in the last place).
408 //
409 // Constants:
410 // The hexadecimal values are the intended ones for the following
411 // constants. The decimal values may be used, provided that the
412 // compiler will convert from decimal to binary accurately enough
413 // to produce the hexadecimal values shown.
414 //
415 // Note: Assuming log() return accurate answer, the following
416 // algorithm can be used to compute log1p(x) to within a few ULP:
417 //
418 // u = 1+x;
419 // if (u==1.0) return x ; else
420 // return log(u)*(x/(u-1.0));
421 //
422 // See HP-15C Advanced Functions Handbook, p.193.
423 //
424 const LN2_HI = kMath[34];
425 const LN2_LO = kMath[35];
426 const TWO54 = kMath[36];
427 const TWO_THIRD = kMath[37];
428 macro KLOGP1(x)
429 (kMath[38+x])
430 endmacro
431
432 function MathLog1p(x) {
433 x = x * 1; // Convert to number.
434 var hx = %_DoubleHi(x);
435 var ax = hx & 0x7fffffff;
436 var k = 1;
437 var f = x;
438 var hu = 1;
439 var c = 0;
440 var u = x;
441
442 if (hx < 0x3fda827a) {
443 // x < 0.41422
444 if (ax >= 0x3ff00000) { // |x| >= 1
445 if (x === -1) {
446 return -INFINITY; // log1p(-1) = -inf
447 } else {
448 return NAN; // log1p(x<-1) = NaN
449 }
450 } else if (ax < 0x3c900000) {
451 // For |x| < 2^-54 we can return x.
452 return x;
453 } else if (ax < 0x3e200000) {
454 // For |x| < 2^-29 we can use a simple two-term Taylor series.
455 return x - x * x * 0.5;
456 }
457
458 if ((hx > 0) || (hx <= -0x402D413D)) { // (int) 0xbfd2bec3 = -0x402d413d
459 // -.2929 < x < 0.41422
460 k = 0;
461 }
462 }
463
464 // Handle Infinity and NAN
465 if (hx >= 0x7ff00000) return x;
466
467 if (k !== 0) {
468 if (hx < 0x43400000) {
469 // x < 2^53
470 u = 1 + x;
471 hu = %_DoubleHi(u);
472 k = (hu >> 20) - 1023;
473 c = (k > 0) ? 1 - (u - x) : x - (u - 1);
474 c = c / u;
475 } else {
476 hu = %_DoubleHi(u);
477 k = (hu >> 20) - 1023;
478 }
479 hu = hu & 0xfffff;
480 if (hu < 0x6a09e) {
481 u = %_ConstructDouble(hu | 0x3ff00000, %_DoubleLo(u)); // Normalize u.
482 } else {
483 ++k;
484 u = %_ConstructDouble(hu | 0x3fe00000, %_DoubleLo(u)); // Normalize u/2.
485 hu = (0x00100000 - hu) >> 2;
486 }
487 f = u - 1;
488 }
489
490 var hfsq = 0.5 * f * f;
491 if (hu === 0) {
492 // |f| < 2^-20;
493 if (f === 0) {
494 if (k === 0) {
495 return 0.0;
496 } else {
497 return k * LN2_HI + (c + k * LN2_LO);
498 }
499 }
500 var R = hfsq * (1 - TWO_THIRD * f);
501 if (k === 0) {
502 return f - R;
503 } else {
504 return k * LN2_HI - ((R - (k * LN2_LO + c)) - f);
505 }
506 }
507
508 var s = f / (2 + f);
509 var z = s * s;
510 var R = z * (KLOGP1(0) + z * (KLOGP1(1) + z *
511 (KLOGP1(2) + z * (KLOGP1(3) + z *
512 (KLOGP1(4) + z * (KLOGP1(5) + z * KLOGP1(6)))))));
513 if (k === 0) {
514 return f - (hfsq - s * (hfsq + R));
515 } else {
516 return k * LN2_HI - ((hfsq - (s * (hfsq + R) + (k * LN2_LO + c))) - f);
517 }
518 }
OLDNEW
« no previous file with comments | « third_party/fdlibm/fdlibm.cc ('k') | tools/generate-runtime-tests.py » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698