christoph/fail
1
Fork 0
Code Releases Activity

perf-test: further results added

Browse Source 
git-svn-id: https://www4.informatik.uni-erlangen.de/i4svn/danceos/trunk/devel/fail@1895 8c4709b5-6ec9-48aa-a5cd-a96041d1645a
This commit is contained in:
adrian
2012-11-08 12:59:43 +00:00
parent 2eb18c81ac
commit e337fef205
1 changed files with 19 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
src/experiments/perf-test/results.txt
View File

@ -71,6 +71,22 @@ perf-test (r1745): Best- vs. Worst-Case with Wallclock-Timer (NON_BP_COUNT = 50
Case A: 321.594s (= ~5min)
Case B: 587.698s (= ~9min)
----------------------------------------------
(E) WITH FAST_BREAKPOINTS AND FAST_WATCHPOINTS (Default mode):
perf-test (r1768 (= ~r1745)): Best- vs. Worst-Case with Wallclock-Timer (NON_BP_COUNT = 50 and BP_COUNT = 50):
Case A: 276.986s (= ~4,6min, around 1,8 times faster than (A).a)
Case B: 538.164s (= ~9,0min, around 8,8 times faster than (A).b)
----------------------------------------------
(F) WITH FAST_BREAKPOINTS AND FAST_WATCHPOINTS (Release mode):
perf-test (r1768 (= ~r1745)): Best- vs. Worst-Case with Wallclock-Timer (NON_BP_COUNT = 50 and BP_COUNT = 50):
Case A: 50.7299s (around 1,2 times slower than (C).a)
Case B: 109.589s (= ~1,8min, around 3,5 times faster than (C).b)
****************************************************************************************************
EVALUATION:
****************************************************************************************************
@ -95,9 +111,12 @@ Note: The following observations and conjectures are partly derived from the ana
callgrind profile (using kcachegrind).
(i) gather() should be inlined. (At the moment, this avoids an include cycle.)
==> impossible (?)
(ii) Bypass the construction of a ResultSet object (the bypass would avoid an additional iteration
over the elements stored in the ResultSet itself), by calling makeActive in gather()
==> declined (!)
(iii) Complete the implementation of the PerfVecSortedSingleBP class (uses binary search in IPs)
==> promising (?)
=> (i) won't effect the speed in Default and Debug mode. (ii) should enable a speedup in all
cases. (iii) will only improve the speed when many *BPSingleListeners* are in use.