Bug 1380101

I tried building linbox, but it fails when compiling tests:

In file included from /usr/include/fflas-ffpack/fflas/fflas_freduce.inl:38:0,
                 from /usr/include/fflas-ffpack/fflas/fflas_freduce.h:56,
                 from /usr/include/fflas-ffpack/fflas/fflas.h:104,
                 from /usr/include/fflas-ffpack/ffpack/ffpack.h:46,
                 from ../linbox/matrix/matrixdomain/blas-matrix-domain.h:45,
                 from ../linbox/matrix/matrix-domain.h:68,
                 from ../linbox/matrix/sparsematrix/sparse-generic.h:80,
                 from ../linbox/matrix/sparse-matrix.h:70,
                 from test-solve.C:45:
/usr/include/fflas-ffpack/utils/bit_manipulation.h: In function 'uint64_t divide_128(uint64_t, uint64_t, uint64_t, uint64_t*)':
/usr/include/fflas-ffpack/utils/bit_manipulation.h:114:10: error: impossible constraint in 'asm'
         );
          ^
This seems to be an error in fflas-ffpack, it seems to assume that amd64 register names are available.

http://koji.fedoraproject.org/koji/taskinfo?taskID=15855078

It built on armv7hl before and the tests passed. Maybe it just doesn't know about aarch64 and assumes x86?

Also, it looks like it's being built with -O2 in CFLAGS:

g++ -DHAVE_CONFIG_H -I. -I..  -DLinBoxTestOnly  -Wall -g -DNDEBUG -U_LB_DEBUG -g -DNDEBUG -U_LB_DEBUG -I../linbox -Wall -DNDEBUG -UFFLASFFPACK_DEBUG -D__FFLASFFPACK_HAVE_CBLAS  -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -fno-strict-aliasing -fopenmp  -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -I/usr/include/atlas      -I/usr/include/m4rie -I/usr/include/saclib  -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1  -c -o test-rational-solver.o test-rational-solver.C
In file included from /usr/include/bits/libc-header-start.h:33:0,
                 from /usr/include/stdint.h:26,
                 from /usr/lib/gcc/aarch64-redhat-linux/6.2.1/include/stdint.h:9,
                 from /usr/include/givaro/givconfig.h:86,
                 from /usr/include/fflas-ffpack/fflas-ffpack-config.h:119,
                 from ../linbox/linbox-config.h:30,
                 from test-rational-solver.C:31:
/usr/include/features.h:360:4: warning: #warning _FORTIFY_SOURCE requires compiling with optimization (-O) [-Wcpp]
 #  warning _FORTIFY_SOURCE requires compiling with optimization (-O)
    ^~~~~~~

... I mean *without* -O2.

... but that's on purpose:

# Don't try to optimize the tests; the build takes gargantuan amounts of memory
sed -i 's|-O2||g' tests/Makefile

It's rather a bug in fflas-ffpack:
fflas_ffpack-2.2.2/fflas-ffpack/utils/bit_manipulation.h:
#ifdef __FFLASFFPACK_HAVE_INT128
// division 128bits by 64 bits
// int128_t(u1,u0) = u1*2^64+u0, div v, rem r
// return quo
static uint64_t divide_128(uint64_t u1, uint64_t u0, uint64_t v, uint64_t *r)
{
        // u0 -> rax
        // u1 -> rdx
        // divq
        uint64_t result;
        __asm__("divq %[v]"
                        : "=a"(result), "=d"(*r)
                        : [v] "r"(v), "a"(u0), "d"(u1)
               );
        return result;
}
#endif

The use of x86 assembly should be guarded by appropriate #ifdefs or intrinsics should be used.

It looks like the Debian maintainer reported a related bug already: https://github.com/linbox-team/fflas-ffpack/issues/45

However, enabling -O optimization for the tests makes the build complete and tests pass on aarch64:

http://koji.fedoraproject.org/koji/taskinfo?taskID=15859942

No issues on other primary arches either, so committed, pushed and new build submitted: http://koji.fedoraproject.org/koji/taskinfo?taskID=15861362 .