Messages in this thread | | | Date | Thu, 26 Jan 2012 17:02:04 -0500 (EST) | Subject | Re: sparc32 vs lib/mpi | From | David Miller <> |
| |
From: Al Viro <viro@ZenIV.linux.org.uk> Date: Sat, 21 Jan 2012 01:12:43 +0000
> I realize that very few people care about sparc32 these days, > but... > a) __clz_tab[] is defined (with identical contents) in > lib/mpi/mpi-bit.c and arch/sparc/lib/divdi3.S. Linker is unhappy, > of course... > b) the same thing ends up using __udiv_qrnnd(), which simply > does not exist on sparc32. There's a Cthulhu-scaring cascade of > ifdefs in lib/mpi/longlong.h, with several variants of udiv_qrnnd() > for sparc. Unfortunately, it falls back on use of __udiv_qrnnd()... > Note that arch/sparc/math-emu/sfp-util_32.h has a definition of > udiv_qrnnd() which might match one of inline assembly variants in > lib/mpi/longlong.h. Or not. BTW, what the hell is LONGLONG_STANDALONE? > > Could somebody who might remember still remember that stuff > come up with comments? For now I'm disabling everything from > INTEGRITY_SIGNATURE down to MPILIB on sparc32 cross-build, but it > would be nice to get it at least somewhat sorted out...
I'm taking a look at this.
| |