Expand description
The literal tables of Chapter 11.11 (private). Every double-double entry
is the pair (RN(c), RN(c โ RN(c))) of the value c named in the
comment of its table, as the bits (hi, lo); every f64 is the value
named, rounded to nearest. They were generated with dashu-float at
2 400 bits and are checked entry by entry, bit for bit, by the unit test
tables_are_the_nearest_double_doubles of the parent module against two
independent oracles.
Constantsยง
- ATAN_C ๐
atan(c)forc = 2^e (1 + (2i + 1)/32),e = โ6, โฆ, 2,i = 0, โฆ, 15(index16(e + 6) + i; Algorithm T).- EXP2_64 ๐
2^(j/64),j = 0, โฆ, 63(Algorithm E).- EXP2_
64_ ๐M1 2^(j/64) โ 1,j = 0, โฆ, 63(Algorithm M, large arguments).- EXPM1_
64 ๐ e^(N/64) โ 1,N = โ31, โฆ, 31(indexN + 31; Algorithm M, small arguments).- FRAC_
3PI_ ๐4 3ฯ/4(the special caseatan2(ยฑโ, โโ)).- INV_
FACT ๐ 1/n!,n = 0, โฆ, 33.- INV_INT ๐
1/n,n = 1, โฆ, 41(entry 0 unused).- INV_
LN2_ ๐64 RN(64 / ln 2): only chooses the indexNof Algorithm E (any nearby value works).- LN2_64 ๐
ln 2 / 64in three parts of 53 bits:RN(c),RN(c โ pโ),RN(c โ pโ โ pโ)(Algorithm E).- LN2_
SPLIT ๐ ln 2in three parts:Arounded to 42 bits (so thateยทAis exact for|e| < 2ยนยน), thenRN(ln 2 โ A),RN(ln 2 โ A โ B)(Algorithm L, step L6).- LN_2C ๐
ln(2(1 + j/64)),j = โ18, โฆ, 0(indexj + 18; Algorithm L,e = 1,y < 1).- LN_C ๐
ln(1 + j/64),j = โ18, โฆ, 26(indexj + 18; Algorithm L).- LN_C_2 ๐
ln((1 + j/64)/2),j = 0, โฆ, 26(Algorithm L,e = โ1,y โฅ 1).- PI_
OVER_ ๐4 โ2^320 ยท ฯ/4โ, big-endian 64-bit words (Algorithm R).- TWO_
OVER_ ๐PI โ2^1408 ยท 2/ฯโ, big-endian 64-bit words: bitiof2/ฯ(weight2^โi) is bit1408 โ iof the integer (Algorithm R).