[Pkg-julia-devel] dsfmt ftbfs on ppc64el

[Graham Inggs]

On 6 November 2015 at 18:56, Peter Colberg <peter at colberg.org> wrote:
> Thanks! The test fails on ppc64el with output check NG! (Japanese for
> No Good), but succeeds on ppc64. I am guessing the Altivec support
> needs to be adapted for little endianness. Fortunately the code is
> is pure C and contains no assembly.

I commented out lines 11-13 from debain/rules and the package built
and passed the tests.

#ifneq (,$(filter $(DEB_HOST_ARCH),ppc64 ppc64el))
#CONFIGURE_FLAGS += --enable-altivec
#endif

On 6 November 2015 at 19:02, Peter Colberg <peter at colberg.org> wrote:
> If you get a chance, please log the performance on ppc64el with
> and without Altivec. The benchmark is run with `test-M19937 -s`.

Results follow.

Regards
Graham


with --enable-altivec:

(sid_ppc64el-dchroot)ginggs at plummer:~/dsfmt/dsfmt-2.2.3+dfsg$ ./test-M19937 -s
consumed time for generating 100000000 randoms.
ST BLOCK [0, 1) AVE: 192ms.
ST BLOCK (0, 1] AVE: 192ms.
ST BLOCK (0, 1) AVE: 208ms.
ST BLOCK [1, 2) AVE: 129ms.
ST SEQ [0, 1) 1 AVE: 302ms.
ST SEQ [0, 1) 2 AVE: 282ms.
total = -nan
ST SEQ (0, 1] 1 AVE: 302ms.
ST SEQ (0, 1] 2 AVE: 282ms.
total = -nan
ST SEQ (0, 1) 1 AVE: 312ms.
ST SEQ (0, 1) 2 AVE: 298ms.
total = -nan
ST SEQ [1, 2) 1 AVE: 295ms.
ST SEQ [1, 2) 2 AVE: 278ms.
total = -nan


without --enable-altivec:

(sid_ppc64el-dchroot)ginggs at plummer:~/dsfmt/dsfmt-2.2.3+dfsg$ ./test-M19937 -s
consumed time for generating 100000000 randoms.
ST BLOCK [0, 1) AVE: 160ms.
ST BLOCK (0, 1] AVE: 159ms.
ST BLOCK (0, 1) AVE: 152ms.
ST BLOCK [1, 2) AVE: 139ms.
ST SEQ [0, 1) 1 AVE: 304ms.
ST SEQ [0, 1) 2 AVE: 284ms.
total = 500014655.815776
ST SEQ (0, 1] 1 AVE: 304ms.
ST SEQ (0, 1] 2 AVE: 284ms.
total = 500035344.184224
ST SEQ (0, 1) 1 AVE: 314ms.
ST SEQ (0, 1) 2 AVE: 301ms.
total = 500014655.815776
ST SEQ [1, 2) 1 AVE: 297ms.
ST SEQ [1, 2) 2 AVE: 280ms.
total = 1500064655.815183