Bug 19732 – make std.complex.expi fast.

std.math.expi is deprecated, one is advised to use std.complex.expi instead. But, unlike the former, std.complex.expi does not make use of fsincos[1] when available. (And it's documentation refers to deprecated std.math.expi for a faster implementation, which is a bit silly.) [1] https://github.com/dlang/phobos/blob/d28a879fecbb7b478e660541bc94c402d38c913d/std/math.d#L3214

Related documentation issues: https://issues.dlang.org/show_bug.cgi?id=19733

Nowadays, the x87 instruction fsincos is slower, not faster, than computing sin and cos separately. https://stackoverflow.com/questions/12485190/calling-fsincos-instruction-in-llvm-slower-than-calling-libc-sin-cos-functions Therefore, there is no need for assembly in the expi() implementation, which was removed in https://github.com/dlang/phobos/commit/029440c2f9008ff2fe8803c97900394bfe9b4e4e.

(In reply to Bastiaan Veelo from comment #2) > Nowadays, the x87 instruction fsincos is slower, not faster, than computing > sin and cos separately. > > https://stackoverflow.com/questions/12485190/calling-fsincos-instruction-in- > llvm-slower-than-calling-libc-sin-cos-functions Have you actually tested this on your box? I recently came to completely different results with LDC on my i5-3550: https://github.com/ldc-developers/ldc/pull/2855#issuecomment-491560143 Note that `long double` is extremely rare in C(++), but ubiquitous in D.

(In reply to kinke from comment #3) > (In reply to Bastiaan Veelo from comment #2) > > Nowadays, the x87 instruction fsincos is slower, not faster, than computing > > sin and cos separately. > > > > https://stackoverflow.com/questions/12485190/calling-fsincos-instruction-in- > > llvm-slower-than-calling-libc-sin-cos-functions > > Have you actually tested this on your box? I recently came to completely > different results with LDC on my i5-3550: > https://github.com/ldc-developers/ldc/pull/2855#issuecomment-491560143 > > Note that `long double` is extremely rare in C(++), but ubiquitous in D. No I have not. What I'd like to know is why fsincos was removed from std.complex.expi in 2012, and why it wasn't readed when std.math.expi was deprecated last year. Feel free to reopen, as you have done the test.

Maybe there should be a std.math.sincos that does not depend on complex numbers at all?