Re:Intel Underestimates Error Bounds by 1.3 Quintillion
by bsder on Thursday October 09, 2014 @04:34PM
(#8434895)
The lesson is actually "This was designed 35+ years ago in different constraints."
Memory and ROM were small and expensive and processors were in-order and slow. Putting the transcendental in microcode made sense, back then.
Now, memory is huge; processors are superscalar and speculative execution; vector units exist; and everybody ignores the built-in transcendentals, but Intel can't dump them for backwards compatibility.
Re:Intel Underestimates Error Bounds by 1.3 Quintillion
by msie on Thursday October 09, 2014 @10:15PM
(#8435918)
Yes, but from the text it seemed like fsin was broken to begin with:
It is surprising that fsin is so inaccurate. I could perhaps forgive it for being inaccurate for extremely large inputs (which it is) but it is hard to forgive it for being so inaccurate on pi which is, ultimately, a very ‘normal’ input to the sin() function.
Re:Intel Underestimates Error Bounds by 1.3 Quintillion
by thaumasiotes on Friday October 10, 2014 @12:08AM
(#8436223)
It's hard for the constraints of old to justify lying about your accuracy in the documentation you produce, though...?
Edit:
The intel engineer was probably smart, but then there's this guy. And then there's the commenters here. Lesson: There's always someone out there smarter than you.