93 lines
3.1 KiB
Plaintext
93 lines
3.1 KiB
Plaintext
# --- db.1.85
|
|
o32:
|
|
30.231u 77.927s 2:30.62 71.8% 0Kshar+0Kunshar+2408Kmax 2085i/434o 49pf+0swaps+9294rec
|
|
|
|
n32:
|
|
28.795u 79.501s 2:28.95 72.7% 0Kshar+0Kunshar+2408Kmax 2207i/302o 128pf+0swaps+9337rec
|
|
35.595u 81.152s 2:21.13 82.7% 0Kshar+0Kunshar+2648Kmax 1990i/243o 53pf+0swaps+11197rec
|
|
|
|
64:
|
|
30.003u 78.765s 2:39.37 68.2% 0Kshar+0Kunshar+2408Kmax 2149i/222o 69pf+0swaps+9588rec
|
|
|
|
# --- I experience a big degradation in performance, in 1.86 vs. 1.85
|
|
|
|
% make (-O3 -IPA -shared)
|
|
% time make test_n32
|
|
90.339u 134.561s 4:26.32 84.4% 0Kshar+0Kunshar+4656Kmax 1760i/880o 709pf+0swaps+31916rec
|
|
|
|
% make (-O2 -OPT:Olimit=0)
|
|
% time make test_64
|
|
51.972u 142.857s 4:17.06 75.7% 0Kshar+0Kunshar+3584Kmax 1164i/709o 103pf+0swaps+11499rec
|
|
|
|
% time make test_o32
|
|
45.108u 123.182s 3:16.81 85.5% 0Kshar+0Kunshar+864Kmax 909i/899o 49pf+0swaps+9992rec
|
|
|
|
% time make test_n32
|
|
42.148u 120.913s 3:04.62 88.3% 0Kshar+0Kunshar+2384Kmax 1237i/530o 317pf+0swaps+10562rec
|
|
46.432u 128.083s 3:36.09 80.7% 0Kshar+0Kunshar+2720Kmax 1590i/541o 400pf+0swaps+10561rec
|
|
|
|
#-----------------------------------------------------------------------------
|
|
# On info2 (4x195MHz CPU, Origin 2000)
|
|
# time make check_n32
|
|
# db.1.86
|
|
hash4/-O3 as the default:
|
|
19.651u 47.540s 1:36.14 69.8% 0Kshar+0Kunshar+864Kmax 1081i/690o 176pf+0swaps+10321rec
|
|
hash5/-O2 as the default:
|
|
19.634u 47.344s 1:33.18 71.8% 0Kshar+0Kunshar+864Kmax 866i/739o 2pf+0swaps+10321rec
|
|
hash5/-O3
|
|
21.328u 47.084s 1:35.52 71.6% 0Kshar+0Kunshar+3544Kmax 872i/785o 5pf+0swaps+11242rec
|
|
hash5/o32/-O3
|
|
21.712u 47.153s 1:57.86 58.4% 0Kshar+0Kunshar+1192Kmax 873i/857o 3pf+0swaps+10561rec
|
|
hash2/n32/-O2
|
|
19.751u 47.492s 1:35.89 70.1% 0Kshar+0Kunshar+864Kmax 748i/761o 3pf+0swaps+10545rec
|
|
|
|
Ouch, all the above were -mips4, we need to use -mips3 since we need to
|
|
Run on Indys/Indigo^2s etc.
|
|
hash5/-O2/mips3
|
|
19.988u 47.326s 1:45.49 63.7% 0Kshar+0Kunshar+864Kmax 866i/816o 2pf+0swaps+10360rec
|
|
|
|
# --- Back to 1.85:
|
|
hash5/-O2/n32/mips3
|
|
14.248u 29.737s 1:05.51 67.1% 0Kshar+0Kunshar+1456Kmax 1994i/221o 16pf+0swaps+11290rec
|
|
|
|
|
|
******************************************************************************
|
|
--- info.engr (O2000) Kudzu, time run.test
|
|
--- statically linked, -n32 -mips3 -O3 -OPT:Olimit=0:space=on, run.test
|
|
|
|
1.85
|
|
----
|
|
libdb/n32/test> time ./run.test
|
|
12.605u 29.758s 0:54.00 78.4% 0+0k 1932+76io 0pf+0w
|
|
|
|
1.86
|
|
----
|
|
libdb/n32/test> time ./run.test
|
|
19.328u 48.854s 1:49.01 62.5% 0+0k 867+705io 0pf+0w
|
|
|
|
Ratios 1.86 vs. 1.85:
|
|
user time: 53% slower
|
|
system time: 64% slower
|
|
|
|
dbhash alone:
|
|
% time ./dbtest86 hash t2hash.inp > /dev/null
|
|
0.023u 0.121s 0:00.22 63.6% 0+0k 3+4io 0pf+0w
|
|
|
|
% time ./dbtest85 hash t2hash.inp > /dev/null
|
|
0.008u 0.012s 0:00.01 100.0% 0+0k 5+0io 0pf+0w
|
|
|
|
So user time is 3x slower, system time is 10x slower.
|
|
I see no difference in dbtree times, only in dbhash times.
|
|
|
|
The speed difference in recno is big as well:
|
|
% time ./dbtest86 recno t4recno.inp > /dev/null
|
|
0.243u 6.682s 0:06.97 99.2% 0+0k 0+0io 0pf+0w
|
|
|
|
% time ./dbtest85 recno t4recno.inp > /dev/null
|
|
0.175u 0.635s 0:00.81 98.7% 0+0k 0+0io 0pf+0w
|
|
|
|
recno 1.86 is 38% slower in user times and again over 10x
|
|
slower in system time.
|
|
|
|
|