SGI GMP

[Roman Marcinek]

Hi,

I think I might have found a problem in SGI/Irix build of the gmp 3.1.1 
and I haven't found anything about it on Your WWW home page so I think it
might be worthwhile reporting it.
   The machine is an old Indy/R4400 running still under Irix 6.2 but with
relativelly new compilers - native compilers ver. 7.2 (higher releases
work only under Irix 6.5, I believe, and I am short in RAM to install that
beast :)
   So the 3.1.1 ver. of gmp nicely compiles with a new ABI (-n32) giving
me both static and shared libraries but any attempt to compile it with an
old ABI (-32 aka -o32) stops short while compiling some of the mpn/*.s
files. The full list of the offending files is:

add_n.s, sub_n.s, addmul_1.s, mul_1.s, submul_1.s, lshift.s, rshift.s

and the compiler message is like the one below:

as0: Error: lshift.s, line 49:Should be even gp register: operand 1

Those files are apparently links from the mips3 subdirectory. So I just
tried to remove those offending files and link them from mips2
subdir which worked for those files. Unfortunatelly later on the compiler
starts to produce complains about lshifts in many of the files, as in the
following:

cc -DHAVE_CONFIG_H -I. -I. -I.. -I.. -DOPERATION_mul_n -O -c mul_n.c -DPIC \
-o .libs/mul_n.lo
cfe: Warning 848: mul_n.c, line 915: Lshift with undefined behaviour.
   const mp_limb_t maskOffHalf = (~(mp_limb_t) 0) << (64  >> 1);
 ---------------------------------^

and whreas those seem to be only warnings and the compiler goes on and at
the end produces the libraries most of the tasts fail damping a core. The
testing programs also exhibit similar lshift problems while being compiled
but I cannot tell if it is this or some other problem. And surely under
-n32 ABI there are no such complains. And I also remember having the same
lshift problems with an older (2.0.2) ver. of the gmp.

   Below I put just selected output lines from the test programs if that
might be any help in solving the problem:

gmake[3]: Entering directory `/usr/people/tmp/gmp-3.1.1/mpz/tests'
ERROR: incorrect square product
multiplier = -7(and some LONG hexadecimal number)
multiplicand = -7(same)
product = F(same)
ref_product = F(also but probably wrong)
Abort - core dumped
FAIL: t-mul
ERROR
dividend = 7F(as above)
divisor  = 70(as above)
Abort - core dumped
FAIL: t-tdiv
ERROR
dividend = -7F...
divisor  = 906E8B5A
Abort - core dumped
FAIL: t-tdiv_ui
ERROR
dividend = -7F...
divisor  = 7E0000007E00000FFFFFFF0FFFFFFF
Abort - core dumped
FAIL: t-fdiv
ERROR
dividend = -7F...
divisor  = 906E8B5A
Abort - core dumped
FAIL: t-fdiv_ui
ERROR
op1 = -FF...
op2 = 7F...
Abort - core dumped
FAIL: t-gcd
ERROR
ref = -1F...
wrong = -1F...
Abort - core dumped
FAIL: dive
ERROR
x2        = 7FFFFFFF7FFFFFFFFFC003FFFFC003FF
x         = F68FFFFFF68FFFFF
remainder = 9286EF019286EF01ECE003FEECE003FE
Abort - core dumped
FAIL: t-sqrtrem
ERROR
op1  = 
-FFFF0000FFFFFF000000FF00000000007FFF00007FFFF80007FCF80007FC0000000000000000
base = 36
Abort - core dumped
FAIL: convert
ERROR
op1  = 
-7FFFFFFF7FFFFFFFFFC003FFFFC003FFF8000FFFF8000FFFFFFFFFFFFFFFFFFFFFC00000FFC000003FFFFFFF3FFFFFFF
op2  = -1000000012E8306802E830680
base = 31
Abort - core dumped
FAIL: io
Abort - core dumped
FAIL: logic
Bus error - core dumped
FAIL: bit
Abort - core dumped
FAIL: t-powm
Abort - core dumped
FAIL: t-powm_ui
Bus error - core dumped
FAIL: t-pow_ui
Bus error - core dumped
FAIL: t-2exp

and here it just ceases go do anything :(

Best regards
dr Roman Marcinek
IF UJ, Krakow, Poland