This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/34265] Missed optimizations
- From: "dominiq at lps dot ens dot fr" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 28 Nov 2007 22:35:49 -0000
- Subject: [Bug tree-optimization/34265] Missed optimizations
- References: <bug-34265-12313@http.gcc.gnu.org/bugzilla/>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Comment #11 from dominiq at lps dot ens dot fr 2007-11-28 22:35 -------
Here are the timings before and after the patch for the polyhedron tests and
some variants:
Before patch After patch
Benchmark Ave Run Number Estim : Ave Run Number Estim
Name (secs) Repeats Err % : (secs) Repeats Err %
--------- ------- ------- ------ : ------- ------- ------
ac 16.92 5 0.0183 : 16.16 5 0.0056
aermod 36.82 5 0.0082 : 36.92 5 0.0106
air 11.38 10 0.0479 : 11.43 11 0.0494
capacita 62.21 5 0.0036 : 61.97 5 0.0343
channel 4.04 12 0.0333 : 4.04 5 0.0160
doduc 58.07 5 0.0257 : 57.56 5 0.0164
fatigue 14.94 5 0.0338 : 14.33 5 0.0184
gas_dyn 11.78 17 0.0448 : 11.89 18 0.0349
induct 93.27 5 0.0093 : 36.93 5 0.0205
linpk 28.15 5 0.0099 : 28.21 5 0.0259
mdbx 16.80 5 0.0112 : 16.83 5 0.0051
nf 32.45 5 0.0388 : 32.63 10 0.0495
protein 55.63 5 0.0069 : 54.86 5 0.0305
rnflow 45.88 5 0.0366 : 46.06 5 0.0230
test_fpu 14.64 5 0.0115 : 14.46 5 0.0207
tfft 3.04 5 0.0380 : 3.06 20 0.0284
ac_v1 16.15 5 0.0197 : 15.16 5 0.0109
air_v1 10.81 5 0.0411 : 10.88 10 0.0471
capacita_8 69.45 5 0.0136 : 69.41 5 0.0091
capacita_10 113.20 5 0.0200 : 112.44 5 0.0290
chan_v1 2.23 5 0.0183 : 2.24 7 0.0471
channel_10 16.61 5 0.0351 : 16.64 14 0.0492
fatigue_v1 13.26 5 0.0071 : 12.05 5 0.0148
fatigue_10 20.54 15 0.0312 : 21.73 5 0.0117
induct_v2 35.07 5 0.0007 : 60.08 5 0.0236
induct_v3 34.40 5 0.0189 : 58.64 5 0.0249
induct_vm 262.95 2 0.0000 : 253.62 2 0.0197
induct_10 100.12 5 0.0053 : 84.65 5 0.0008
kepler 22.73 5 0.0123 : 26.11 5 0.0069
kepler_10 69.59 5 0.0047 : 61.42 5 0.0110
nf_10 58.00 5 0.0413 : 58.36 5 0.0388
protein_10 57.04 5 0.0167 : 56.38 5 0.0486
test_fpu_v1 15.15 5 0.0195 : 14.98 5 0.0104
test_fpu_10 34.75 5 0.0408 : 34.68 5 0.0120
tfft_8 6.81 5 0.0110 : 6.83 5 0.0371
tfft_10 14.36 5 0.0373 : 14.40 6 0.0496
Before patch After patch
Benchmark Compile Executable : Compile Executable
Name (secs) (bytes) : (secs) (bytes)
--------- ------- ---------- : ------- ----------
ac 4.52 50628 : 4.60 50628
aermod 96.22 1288460 : 106.72 1288460
air 6.57 80956 : 6.68 80956
capacita 3.18 60140 : 3.34 64236
channel 1.55 38532 : 1.65 38532
doduc 13.52 183264 : 14.02 191456
fatigue 5.69 84564 : 5.83 80468
gas_dyn 5.36 695776 : 5.50 695776
induct 11.65 160132 : 12.02 168324
linpk 1.67 46512 : 1.71 46512
mdbx 3.76 72672 : 3.85 72672
nf 4.45 87644 : 4.46 87644
protein 11.18 113900 : 11.45 113900
rnflow 11.58 187316 : 11.74 187316
test_fpu 11.23 182544 : 11.09 178448
tfft 1.30 34420 : 1.33 34420
ac_v1 4.51 50628 : 4.60 50628
air_v1 6.63 80956 : 6.75 80956
capacita_8 3.20 60136 : 3.33 64232
capacita_10 3.19 64216 : 3.42 68312
chan_v1 1.85 38500 : 1.85 38500
channel_10 1.32 34392 : 1.40 34392
fatigue_v1 5.75 84524 : 5.77 80428
fatigue_10 4.91 76352 : 4.91 76352
induct_v2 11.72 168324 : 12.19 172420
induct_v3 11.72 164228 : 12.12 172420
induct_vm 11.44 160132 : 11.73 164228
induct_10 11.47 159964 : 11.89 159964
kepler 0.34 17652 : 0.35 17652
kepler_10 0.33 17632 : 0.34 17632
nf_10 2.03 46684 : 2.07 46684
protein_10 7.01 93400 : 7.18 93400
test_fpu_v1 11.23 182592 : 11.17 178496
test_fpu_10 6.54 117056 : 6.40 117056
tfft_8 1.25 30348 : 1.31 30348
tfft_10 1.15 30328 : 1.18 30328
The only timings significantly changed by the patch are the induct avatars,
with the strange result that the variants which missed the vectorization are
now vectorized, while those previously vectorized are not any more (also true
for the variants of the first attachment). So there is probably some need of a
little bit of tuning.
I have also to regtest and do some further investigations.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34265