This is actually a bit more subtle than you&#39;d think.  Are those constants precise and exact?  (There&#39;s certainly floating point code that exploits the cancellations in the floating point model) There&#39;s many floating point computations that can&#39;t be done with exact rational operations.  There&#39;s also certain aspects that are target dependent like operations having 80bit vs 64bit precision. (Ie using the old intel fp registers vs sse2 and newer)<div>
<br></div><div>What&#39;s the ticket you&#39;re working on?  </div><div><br></div><div><br></div><div><span></span>Please be very cautious with floating point, any changes to the meaning that aren&#39;t communicated by the programs author could leave a haskeller numerical analyst scratching their head.  For example, when doing these floating point computations, what rounding modes will you use? <br>
</div><div><br>On Monday, January 13, 2014, Kyle Van Berendonck  wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">Hi,<div><br></div><div>I&#39;m cutting my teeth on some constant folding for floats in the cmm.</div>
<div><br></div><div>I have a question regarding the ticket I&#39;m tackling:</div><div><br></div><div>Should floats be folded with infinite precision (and later truncated to the platform float size) -- most useful/accurate, or folded with the platform precision, i.e. double, losing accuracy but keeping consistent behaviour with -O0 -- most &quot;correct&quot;?</div>

<div><br></div><div>I would prefer the first case because it&#39;s *much* easier to implement than the second, and it&#39;ll probably rot less.</div><div><br></div><div>Regards.</div></div>
</blockquote></div>