[PATCH 2/8] test/math: Add missing libm-test-ulps-* files

Bernhard Reutner-Fischer rep.dot.nop at gmail.com
Tue Feb 18 21:43:41 UTC 2014


On Tue, Feb 18, 2014 at 10:36:05PM +0100, Bernhard Reutner-Fischer wrote:
> On Tue, Feb 18, 2014 at 05:47:37PM +0530, Vineet Gupta wrote:
> > On Tuesday 18 February 2014 01:23 PM, Bernhard Reutner-Fischer wrote:
> 
> > And what about the architectures not ported to glibc and hence don't have a
> > libm-test-ulps-* yet.
> > 
> > Is there a way to start off with a stub, touching an empty file doesnt seem to work.
> 
> I don't know. Let's see:
> Read glibc/math/README.libm-test. So in uClibc:
> (obviously make sure to enable XSI and C99 _MATH in your .config)
> make install_dir;# or realclean all install_dir
> cd test/math
> make compile
> # mhm, let's build the generators manually..
> make V=2 test-double
> vi libm-tests.c, look for argv handling;
> # remember what the readme told us:
> rm -f *ULPs*
> # on the target, just one as an example:
> for i in double;do ./test-$i -i yes -u ;mv ULPs ULPs-$i;done
> # above should not spit out too many errors, ideally. Check your impl otherwise
> # back on the build

oops, the readme told us to run the script on all the input at once, so
cat ULPs-* > raw-arc
./gen-libm-test.pl -u raw-arc -n
mv NewUlps libm-test-ulps-arc
rm -f *ULPs* *Ulps* *raw-arc*

> for i in double;do ./gen-libm-test.pl -u ULPs-$i -n;mv NewUlps NewUlps-$i;done
> cat NewUlps-* > libm-test-ulps-arc
> rm -f *ULPs* *Ulps* ;# cleanup
> 
> Do plausibility checks on ALL the gathered data, fix, goto start.
> 
> I guess you get the idea.
> Does that help?


More information about the uClibc mailing list