------- Comment #4 from kkojima at gcc dot gnu dot org 2008-06-11 01:27 ------- Created an attachment (id=15746) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15746&action=view) patch
>From the reports on gcc-testresults, cdce3.C fails on some other targets: arm-none-linux-gnueabi http://gcc.gnu.org/ml/gcc-testresults/2008-06/msg00404.html powerpc64-unknown-linux-gnu http://gcc.gnu.org/ml/gcc-testresults/2008-06/msg00430.html sh4-unknown-linux-gnu http://gcc.gnu.org/ml/gcc-testresults/2008-06/msg00494.html powerpc-ibm-aix5.3.0.0 http://gcc.gnu.org/ml/gcc-testresults/2008-06/msg00508.html i386-unknown-freebsd6.2 http://gcc.gnu.org/ml/gcc-testresults/2008-06/msg00522.html mipsel-unknown-linux-gnu http://gcc.gnu.org/ml/gcc-testresults/2008-06/msg00555.html though now cdce3.C test is skipped on darwin. It fails also for some *-elf embedded targets. For these *-linux* and *-elf targets, it seems that the lack of 'l' suffixed math functions causes errors. How about the attached patch? It tests long double math functions only for the targets having true long double and skips tests for pow10*/exp10* functions which are gnu extensions on *-*-darwin*. It's tested for i686-pc-linux-gnu, powerpc-apple-darwin9.0.0, sh-unknown-elf and sh4-unknown-linux-gnu. I'm not sure how this test fails on AIX and FreeBSD, though. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36440