Results for Exercise Sheet 3 (Intersect)
Add your row to the table below, following the examples already there. Please write down the time in microseconds as an integer, and the ratios as floats with exactly two digits of precision. In the first column, write your first name, or name1+name2 if you work in a group. Speedup is always measured as time_for_baseline / time_for_your_version (<1 = slower, >1 faster). In the last column briefly describe the improvements you implemented.
As mentioned in the lecture, repeat each measurement 5 times and take the average!
Name |
Average time |
existence + bielefeld |
them + bielefeld |
them + existence |
Language |
Machine Spec |
Implementation Summary |
Patrick |
1698µs |
1.00 |
1.00 |
1.00 |
C++ |
Intel i5-3320M @ 2.60GH - 8 GB RAM |
None (Baseline) |
dk295 |
1099µs (baseline 1455µs) |
184/118=1.56 |
1213/745=1.63 |
2968/2433=1.22 |
C++ |
Intel Core i7-5600U CPU @ 2.60GHz - 12 GB RAM |
bin search for big list len diff, zipper else. mult non-algorithmic optim |
jk544 |
1993µs |
1.65 |
2.31 |
1.32 |
C++ |
Intel Core i3-5005U CPU @ 2.00GHz - 4GB RAM |
Galloping search for big len diff, simple skipping of 30 elements for medium diff, non-alg improved zipper for small len diff |
km151 |
8230µs |
1.64 |
1.46 |
17.22 |
Java |
Intel® Core™ i5-2410M CPU @ 2.30GHz × 4 - 6GB RAM |
Binary search kwith predictable branches if just one of the Lists < 10000 else zipper, non-alg improved zipper with Sentinels, predictable branches, native arrays |
kh143 |
2004µs |
1.09 |
7.94 |
3.79 |
Java |
AMD Phenom II X4 955 @ 3.20GHz - 12 GB RAM |
galloping search for lists with length1 < 10000 and length2 > 100000, baseline for lists both < 200000, else zipper with sentinels and native int array |
ts341 |
6187µs |
1.65 |
4.07 |
2.34 |
Java |
Intel Core i5-4210U CPU @ 1.70GHz × 4 - 8GB RAM |
Binary search if list length difference factor > 100, else zipper; sentinels; predictable branches; native arrays. |
mf233 |
1762µs |
1.46 |
1.88 |
1.13 |
C++ |
Intel Celeron CPU G1610 @ 2.6GHz - 4GB RAM |
For short list differences --> improved zipper. For long differences --> combination of binary search and linear search (always skipping 32 elements) |
ff115 |
1876µs |
1.65 |
0.84 |
1.11 |
C++ |
Intel® Core™ i5-3337U CPU @ 1.80GHz × 4 - 4GB RAM |
Galloping search if list length difference factor > 50, else zipper with sentinels |
ms801 |
1356µs |
1.26 |
1.91 |
1.23 |
C++ |
Intel® Core™ i5-3210M CPU @ 2.50GHz × 2 - 8GB RAM |
Zipper with sentinels for medium list difference, else binary search |
jb657 |
6667µs |
1.00 |
1.00 |
5.50 |
C++ |
Intel® Core™ i5-2520M CPU @ 2.50GHz × 2 - 4GB RAM |
if shorter list 100 times shorter than longer list: Galloping search. else: zipper |
bt45 |
1756µs |
2.53 |
1.46 |
1.34 |
Java |
Intel i7-2620M @ 2.7GHz - 8 GB RAM |
if length-ratio < 15 use skipper, else binary with remainders |
sh477 |
1720µs |
1.02 |
4.52 |
2.43 |
Java |
Intel Core i7-6700K CPU @ 4.00GHz - 16 GB RAM |
binary search if difference of list lengths bigger than factor 20, else zipper with sentinels, less if and native arrays |
jg292 |
1220µs(BL: 1547) |
1.19 |
2.12 |
1.08 |
C++ |
Intel(R)Core(TM) i5-4200M CPU @ 2.50GHz 12 GB RAM |
Sentinel, pred. branches, use skipper of 60 elements if length ratio >4 |
ss515 |
926µs(BL: 1138) |
1,03 |
2,57 |
1,00 |
C++ |
Intel(R) Core(TM) i5-4590 CPU @ 3.30GHz 16 GB RAM |
pred. branches, galloping search with shorter list, switch method depending of length difference (factor 50) |
Edit conflict - other version:
pb138 |
5444µs |
1.41 |
2.52 |
1.15 |
Java |
Intel® Core™ i7-3632QM CPU @ 2.20GHz - 4GB RAM |
native arrays, sentinels, if length dif factor 100 bin search, else zipper |
Edit conflict - your version:
pb138 |
5444µs |
1.41 |
2.52 |
1.15 |
java |
Intel® Core™ i7-3632QM CPU @ 2.20GHz - 4GB RAM |
native arrays, sentinels, if length dif factor 100 bin search, else zipper |
pb138 |
1044µS |
0.97 |
2.58 |
2.09 |
C++ |
Intel® Core™ i7-4770 CPU @ 3.40GHz - 24GB RAM |
Use binary search for length differences with factor > 100, else zipper; some compiler flags; zipper is a bit faster with clang, but binary search is much slower than with gcc |
End of edit conflict