Cache friendly strength and degree

[gendelpiekel]

Implements #2599. Specifically:

• faster igraph_degree() when ignoring loops and calculating degree for all vertices,
• faster igraph_strength() when calculating strength for all vertices,
• Added benchmark for igraph_strength().

Local benchmark results

igraph_strength() before:

$ ./tests/benchmark_igraph_strength

|  1 igraph_strength(), preferential attachment n=10000, m=10, 100x                 0.03s   0.03s      0s
|  2 igraph_strength(), preferential attachment n=100000, m=10, 100x                0.44s  0.439s      0s
|  3 igraph_strength(), preferential attachment n=100000, m=100, 10x               0.971s  0.971s      0s

igraph_strength() after:

$ ./tests/benchmark_igraph_strength

|  1 igraph_strength(), preferential attachment n=10000, m=10, 100x                0.015s  0.015s      0s
|  2 igraph_strength(), preferential attachment n=100000, m=10, 100x               0.219s  0.219s      0s
|  3 igraph_strength(), preferential attachment n=100000, m=100, 10x               0.284s  0.284s      0s

igraph_degree() before:

$ ./tests/benchmark_igraph_degree

[...]

Do not count loops, O(d) per degree.
|  1 igraph_degree(), preferential attachment n=100000, m=10, 100x                 0.331s  0.331s      0s

igraph_degree() after:

$ ./tests/benchmark_igraph_degree

[...]

Do not count loops, O(d) per degree.
|  1 igraph_degree(), preferential attachment n=100000, m=10, 100x                 0.197s  0.197s      0s

[szhorvat]

Thank you, this is great work! Are you new to igraph, or have you used it before?

The two comments are just an FYI. I'll take care of these and will merge the PR.

If you'd like to be credited using your real name (in addition to your github handle), please let me know what it is.

[codecov]

Codecov Report

Attention: Patch coverage is 97.56098% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 84.29%. Comparing base (c6606ae) to head (c2a6dfa).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2602      +/-   ##
==========================================
- Coverage   84.33%   84.29%   -0.04%     
==========================================
  Files         379      379              
  Lines       61908    61949      +41     
  Branches    12116    12137      +21     
==========================================
+ Hits        52210    52221      +11     
- Misses       9698     9728      +30     

Files	Coverage Δ
src/graph/type_indexededgelist.c	95.87% <100.00%> (-0.26%)	⬇️
src/properties/degrees.c	80.97% <96.55%> (-9.41%)	⬇️

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c6606ae...c2a6dfa. Read the comment docs.

[szhorvat]

I updated the benchmarks to compare the previous implementation (b) with this faster code path (a), and I'm getting the following:

|> Benchmark file: /Users/szhorvat/Repos/igraph-main/igraph/tests/benchmarks/igraph_strength.c
|  1a igraph_strength(), preferential attachment n=1000, m=10, 10000x               0.43s  0.422s      0s
|  1b igraph_strength(), preferential attachment n=1000, m=10, 10000x               0.26s  0.256s      0s

|  2a igraph_strength(), preferential attachment n=10000, m=10, 1000x              0.397s  0.392s      0s
|  2b igraph_strength(), preferential attachment n=10000, m=10, 1000x               0.28s  0.276s      0s

|  3a igraph_strength(), preferential attachment n=100000, m=10, 100x              0.422s  0.418s      0s
|  3b igraph_strength(), preferential attachment n=100000, m=10, 100x              0.423s  0.417s      0s

|  4a igraph_strength(), preferential attachment n=100000, m=100, 10x              0.427s  0.421s      0s
|  4b igraph_strength(), preferential attachment n=100000, m=100, 10x              0.868s  0.855s  0.001s