[Tinycc-devel] Commit f077d16c (gen_cast: cast FLOAT to DOUBLE)

[Michael Matz]

Hey grischka,

some of my patches interact with $subject, and I have difficulties to 
determine what the purpose of that commit was.  The comments ("to avoid 
precision loss" (... when converting to int)) don't really clarify much. 
I also can't find testcases (on x86-64) where the commit mattered, so 
before starting to work around that in my patch can you please clarify its 
purpose?

(FWIW: the implementation of fix{uns,}sfdi as well as gen_cvt_ftoi on 
x86-64 and arm64 are IMO correct.  Were you perhaps worried about the x87 
registers always internally using wide precision?)


Ciao,
Michael.