long double: drop ld128 support? and move ldshape union to arch/ (at least rename ieeel2 union and use the same union in all *l.c) volatile fix: -ffloat-store or -fexcess-precision=standard probably fixes most volatile issues (and strict_assign) volatile const may need different treatment rounding functions: floor, ceil.. probably can be improved math.h: nan 0/0 raises invalid signbit >>63 or !! exceptions: uniform inexact/invalid exception raising: (int)x == 0 and huge+x > 0 are used (unreachable code is not commented and hard to follow) log2: dekker vs long double arithmetics ldhack.h fix: move ldouble manipulation into arch specific headers long double isnan, isinf..: efficient macro versions of long double classification macros? (they are ld* representation specific) round: no inexact exception (or fix lround) rint: use +0x1.8p52 or asm trigonometric functions: __rem_pio_large: is returning 2 bits enough? __tan: 3rd arg semantics is probably not optimal sinf,cosf: return sindf(-y) vs -sindf(y)? __sin,__cos,..: z,w,.. -> x2,x4,.. so degree is easier to see __rem_pio2*: rounding trick: simply use the low 32 bits of fn? use long double pi in long double code? (casin, cacos,..) scalbf: scalb is buggy, do we need the *f and *l version? generic code fixes: int32_t -> uint32_t conversion (subtle) += 1, -= 1 -> ++, -- TWO52, twom1000 vs tiny remove "unsupported long double format" checks remove overflow thresholds (sinh, cosh) when result overflows anyway? sign bit checking convention (sqrt.c) missing: sqrtl tgamma, tgammaf (long double bessel) nextafterf on ld64 tgamma: lanczos approx as in boost/math/special_functions and python/Modules/mathmodule.c or bsd tgamma complex optimizable creal cimag (libm.h macro for internal code?) include and instead of libm.h (once there are efficient creal etc.) cpack(x,y) vs x+I*y vs union .a[0]=x, .a[1]=y fix casin[h], cacos[h], catan[h] (complex arith cornercases) add missing long double versions Kahan, W. "Branch Cuts for Complex Elementary Functions, or Much Ado About Nothing's Sign Bit." 1987 Hull, Fairgrieve, Tang "Implementing complex elementary functions using exception handling" 1994 Hull, Fairgrieve, Tang "Implementing the complex arcsine and arccosine functions using exception handling" 1997 boost/math/complex python/Modules/cmathmodule.c asm: x86_64 asm versions libm test test vectors (see if python has good math+cmath tests) special cases from c99 F.9, G.6 randomized test using mpfr and mdc for complex udiv: add udiv.s bugfixes: code depends on signed int right shift behaviour pow*(1,nan or inf) == 1 special case nan check in modf fmal ldexp vs ldexpl fma* x == 0 should be after the isfinite check invalid ldexp optimizaion lrint* inexact clearing was optimized away scalbn(0x1p1023, -2097)==0x1p-1074 was incorrect ilogbl was incorrect on ld80 (implicit bit was not taken into account)