← Back to branch summary

~ubuntu-branches/ubuntu/raring/blitz++/raring

« back to all changes in this revision

Viewing changes to benchmarks/ctime-results

  • Committer: Bazaar Package Importer
  • Author(s): Konstantinos Margaritis
  • Date: 2005-02-28 20:25:01 UTC
  • mfrom: (2.1.2 hoary)
  • Revision ID: james.westby@ubuntu.com-20050228202501-3i4f2sknnprsqfhz
Tags: 1:0.8-4
http://bugs.debian.org/297323
Added missing build-depends (Closes: #297323)

Show diffs side-by-side

added added

removed removed

Lines of Context:
benchmarks/ctime-results
 
1
On olympus.extreme.indiana.edu (sparc-sun-solaris2.6):
 
2
egcs 1.1b
 
3
 
 
4
Initial version, with -O2 -ftemplate-depth-30 -O2 -funroll-loops 
 
5
  -fstrict-aliasing 
 
6
 
 
7
ctime1    17.7 0.9
 
8
ctime2    25.7 1.2
 
9
ctime3    52.0 2.1
 
10
ctime4   sleep
 
11
 
 
12
With -fno-gcse:
 
13
ctime1  17.3 1.0
 
14
ctime2  26.3 1.3
 
15
ctime3  1:02.0 2.1
 
16
ctime4  sleep
 
17
 
 
18
With -O:
 
19
ctime1  17.3 0.8
 
20
ctime2  24.4 1.2
 
21
ctime3  51.5 2.1
 
22
ctime4  sleep
 
23
 
 
24
With -O -fno-inline:
 
25
ctime1  16.9 0.8
 
26
ctime2  20.0 1.0
 
27
ctime3  24.7 1.2
 
28
ctime4  31.2 1.6
 
29
 
 
30
Woohoo.  Okay, obviously inlining is the key.
 
31
 
 
32
Now try new expression templates:
 
33
 
 
34
With -O -funroll-loops -DBZ_NEW_EXPRESSION_TEMPLATES
 
35
ctime1  14.1 0.9
 
36
ctime2  22.3 1.2
 
37
ctime3  58.8 2.2
 
38
 
 
39
With -O -funroll-loops -DBZ_NEW_EXPRESSION_TEMPLATES -DBZ_NO_INLINE_ET
 
40
ctime1  14.1 0.9
 
41
ctime2  21.1 1.0
 
42
ctime3  45.4 1.9
 
43
 
 
44
With -O -funroll-loops -DBZ_NEW_EXPRESSION_TEMPLATES -DBZ_NO_INLINE_ET -DBZ_ETPARMS_CONSTREF
 
45
ctime1  14.6 0.8
 
46
ctime2  20.7 1.1
 
47
ctime3  41.6 2.1
 
48
ctime4 1:27.7 3.0
 
49
 
 
50
Things to try:
 
51
-fno-inline
 
52
 
 
53
Just -O (this will turn off -funroll-all-loops)
 
54
-fno-expensive-optimizations
 
55
-fno-unroll-all-loops
 
56
-fno-strength-reduce
 
57
-fno-rerun-cse-after-loop
 
58
 
 
59
 
 
60
 
 
61
 
 
62
 
 
63
 
 
64
On hgar1.cwru.edu (alpha), with KCC:
 
65
 
 
66
With +K3 -O3 -DBZ_NEW_EXPRESSION_TEMPLATES -DBZ_NO_INLINE_ET -DBZ_ETPARMS_CONSTREF:
 
67
ctime1  13.1 0.8
 
68
ctime2  20.9 1.0
 
69
ctime3  27.3 1.0
 
70
ctime4  36.2 1.1
 
71
ctime5  48.7 1.2
 
72
 
 
73
With just +K3 -O3:
 
74
ctime1  15.8 0.9
 
75
ctime2  25.3 1.0
 
76
ctime3  46.2 1.2
 
77
ctime4  79.9 1.5
 
78
 
 
79
So a speed up of about X 2 with KCC, not counting the overhead.
 
80
 
 
81
 
 
82
Here are the results for <valarray>:
 
83
ctime1  0.9 0.2
 
84
ctime2  2.1 0.2
 
85
ctime3  9.4 0.3
 
86
ctime4  33.2 0.4
 
87
ctime5  1:13 0.6
 
88
 
 
89
 
 
90
For C code:
 
91
ctime5  0.35 0.08
 
92
 
 
93
Pretty terrible.
 
94