GMP 4.2 make check failures

[Marc Glisse]

On Thu, 5 Feb 2009, Samuelson, Gary M wrote:

> GMP version 4.2

Is this 4.2 As in 4.2.0? This is a really old version, you should get 
4.2.4, assuming that 4.3.0 is not out yet.

> ld.so.1: t-bswap: fatal: /usr/local/lib/libgcc_s.so.1: wrong ELF class:
> ELFCLASS32

You should really check the archives of the mailing list. gcc comes with 
broken specs by default. Ways to work around it include modifying these 
specs, adding -R/usr/local/lib/sparcv9 to LDFLAGS, defining 
LD_LIBRARY_PATH_64, using crle, etc. But my guess is that you are actually 
more interested in the 32 bit version of gmp, so you could try:
ABI=32
export ABI
./configure ...

Note that the 64 bit version is currently faster if you want to use gmp 
for intensive computation.

(I am a bit surprised that the 32 bit version can't be made almost as 
fast, since from what I understand both have access to the same 
instructions, as I was reminded reading: 
http://www.opensolaris.org/jive/thread.jspa?threadID=90305&tstart=0
but I know the people who wrote the gmp code know much more than I do 
about this, and the 32 bit code already uses 64 bit instructions in some 
places)

> SPROcc          Sun Studio 11 C Compiler
>                 (sparc) 11.0,REV=2005.10.13

Does it have the latest patches?
http://developers.sun.com/sunstudio/downloads/patches/ss11_patches.html

> mpn_get_d wrong on random data
>    sign     0
>    n        =0x1FF800FE007FC0
>    nsize    1
>    exp      -37
>    want     =[43 3F F8 00 FE 00 7F C0] 8998407423164352
>    got      =[40 EF F8 00 FE 00 7F C0] 65472.031006097327918
> FAIL: t-get_d

There, the archives of the mailing list suggest this is caused by an 
unsafe optimization triggered by the -fast flag in Sun Studio. I believe 
replacing -fsimple=1 with -fsimple=0 is enough to work around it, but it 
may be easier to just specify your own CFLAGS that don't include -fast.

-- 
Marc Glisse