Re: [Bug-gnubg] Benchmarks, experiments, speedups

[Øystein Johansen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ingo Macherius wrote:
| Hmmm. The ftree-vectorize is indeed the key flag and should boost
| performance. See:
| http://gcc.gnu.org/projects/tree-ssa/vectorization.html You should
| verify the workings of the vectorizer by running the verbose mode,
| -ftree-vectorizer-verbose=5

Aha!

- -ftree-vectorizer-verbose does reveal something!

Attached output of compilation.

- -Øystein
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (MingW32)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFCOxUt6kDTFPhwyqYRAo1gAJ4n13VmTrQxozIurZbjAVahzuTjtgCfdVmx
EW8x91Rbqw1AbE2ROcfctGI=
=DnlT
-----END PGP SIGNATURE-----

/msys/1.0/local/bin/gcc.exe -mms-bitfields -O3  -ftree-vectorize 
-ftree-vectorizer-verbose=5 -msse2  -Wall -DHAVE_CONFIG_H  -I. -I..   -c -o 
neuralnet.o neuralnet.c

neuralnet.c:323: note: not vectorized: possible dependence between data-refs 
rc.randrsl[0] and rc.randrsl[i_383]
neuralnet.c:876: note: not vectorized: too many BBs in loop.
neuralnet.c:887: note: not vectorized: nested loop.
neuralnet.c:888: note: not vectorized: too many BBs in loop.
neuralnet.c:908: note: not vectorized: too many BBs in loop.
neuralnet.c:919: note: not vectorized: nested loop.
neuralnet.c:920: note: not vectorized: too many BBs in loop.
neuralnet.c:865: note: vectorized 0 loops in function.

neuralnet.c:1041: note: not vectorized: multiple exits.
neuralnet.c:1045: note: not vectorized: multiple exits.
neuralnet.c:1049: note: not vectorized: multiple exits.
neuralnet.c:1053: note: not vectorized: multiple exits.
neuralnet.c:1030: note: vectorized 0 loops in function.

neuralnet.c:323: note: not vectorized: possible dependence between data-refs 
rc.randrsl[0] and rc.randrsl[i_277]
neuralnet.c:373: note: not vectorized: too many BBs in loop.
neuralnet.c:376: note: not vectorized: too many BBs in loop.
neuralnet.c:379: note: not vectorized: too many BBs in loop.
neuralnet.c:382: note: not vectorized: too many BBs in loop.
neuralnet.c:333: note: vectorized 0 loops in function.

neuralnet.c:1021: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:1024: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:988: note: vectorized 0 loops in function.

neuralnet.c:964: note: not vectorized: multiple exits.
neuralnet.c:968: note: not vectorized: multiple exits.
neuralnet.c:972: note: not vectorized: multiple exits.
neuralnet.c:976: note: not vectorized: multiple exits.
neuralnet.c:979: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:982: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:942: note: vectorized 0 loops in function.

neuralnet.c:457: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:462: note: not vectorized: nested loop.
neuralnet.c:472: note: not vectorized: unsupported use in stmt.
neuralnet.c:469: note: not vectorized: unsupported use in stmt.
neuralnet.c:482: note: not vectorized: too many BBs in loop.
neuralnet.c:488: note: not vectorized: nested loop.
neuralnet.c:491: note: not vectorized: unsupported use in stmt.
neuralnet.c:448: note: vectorized 0 loops in function.

neuralnet.c:818: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:823: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:828: note: not vectorized: nested loop.
neuralnet.c:829: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:832: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:838: note: not vectorized: nested loop.
neuralnet.c:839: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:846: note: not vectorized: nested loop.
neuralnet.c:853: note: not vectorized: ptr is loop invariant.
neuralnet.c:853: note: not vectorized: unhandled data ref: D.3815_135 = 
*D.3814_110
neuralnet.c:849: note: not vectorized: can't determine dependence between: 
*prWeight_12 and *prWeight_12
neuralnet.c:856: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:801: note: vectorized 0 loops in function.

neuralnet.c:719: note: not vectorized: nested loop.
neuralnet.c:726: note: not vectorized: number of iterations cannot be computed.
neuralnet.c:702: note: vectorized 0 loops in function.

neuralnet.c:679: note: not vectorized: too many BBs in loop.
neuralnet.c:567: note: not vectorized: nested loop.
neuralnet.c:581: note: not vectorized: unsupported use in stmt.
neuralnet.c:578: note: not vectorized: unsupported use in stmt.
neuralnet.c:574: note: not vectorized: unsupported use in stmt.
neuralnet.c:587: note: not vectorized: too many BBs in loop.
neuralnet.c:593: note: not vectorized: nested loop.
neuralnet.c:596: note: not vectorized: unsupported use in stmt.
neuralnet.c:649: note: vectorized 0 loops in function.
ar cr libevent.a hash.o heap.o list.o neuralnet.o dynarray.o isaac.o md5.o 
mt19937ar.o 
ranlib libevent.a