On Sun, Sep 15, 2024 at 07:53:54PM -0400, Greg Troxel wrote:
> I'm doing all this on netbsd-10, amd64, system gcc 10, intel 9th
> generation i7. The following test program calls remquo with (-90, 90).
> This is from a test in geodesiclib in proj of a high-accuracy sincos
> implementation, which re
Date:Sun, 15 Sep 2024 19:53:54 -0400
From:Greg Troxel
Message-ID:
| C99/POSIX define remquo:
| https://pubs.opengroup.org/onlinepubs/9799919799/functions/remquo.html
I know really nothing about any of this, but that's still the 2018
version of POSIX, and I do
On Sun, 15 Sep 2024, Greg Troxel wrote:
The test program calls remquo twice, one with the -90 in a double, and
once with it as a literal. The first time gets a positive q -- which
messes up the sin calculation, wwhich is how I got here. The second is
correct.
Not a math nerd, but, I can exp
FreeBSD has a change, which resolves the sign of q bug (1 vs -1):
However, it still has q being 0 when it should be -2^n.
The patch at end causes
my remquo test programs
proj's regression tests
to all pass.
The first three hunks are from FreeBSD:
https://cgit.freebsd.org/src/commit/lib/ms
C99/POSIX define remquo:
https://pubs.opengroup.org/onlinepubs/9799919799/functions/remquo.html
The basic idea is that return value r is a remainder, and in q is stored
the low-order k bits of the quotient, but with q having the same sign as
the quotient.
The point is to reduce possibly large
