Performance Benchmarks

[d4tocchini]

I noticed your comment on Greg Badro's blog stating you saw a 2x speed boost from the original js port. Awesome! Do you have any tests that demonstrate this, jsperf or whatever? Did you see a general speed boost or for particular cases? Is the performance boost to just from removing external libraries?

[slightlyoff]

Heya:

I'm still pretty early in the profiling. I've been mostly working to get rid of external dependencies 'till now, but it is getting somewhat faster as I find and kill dumb code that's probably the result of the transliteration (not stupid assumptions in the original...language impedance mismatches do that to you EVERY TIME).

Something like JSPerf is a much better idea than what I've been doing. I'll add it to my TODO list.

The most recent code is ~5-6x faster than the original on the unit tests. Here's some truncated before/after. The headline numbers are 1.78s for the original vs. 0.325 for the refactored code. More to do, though:

---

obelisk:js slightlyoff$ /projects/v8_edge/d8 --harmony CL-loads.js
...
addDel:
starting timing test. nCns = 100, nSolvers = 10, nResolves = 50
done building data structures
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.014
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.027
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.052
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.067
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.082
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.097
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.111
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.124
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.138
done adding 61 constraints [500 attempted, 0 exceptions]
time = 0.155
Editing vars with indices 74, 54
about to start resolves
done resolves -- now ending edits
total time = 1.782

  number of constraints:        100
  number of solvers:            10
  numbers of resolves:          50
  tests:                1
  time to add (ms):             155
  time to edit (ms):            23
  time to resolve (ms):         641
  time to edit (ms):            13
  add time per solver (ms):         0.155
  edit time per solver (ms):        1.15
  resolve time per resolve (ms):    1.282
  time to end edits per solver (ms):    0.65

---

obelisk:cassowary-js-refactor slightlyoff$ /projects/v8_edge/d8 --harmony run-cl-tests.js 
...
addDel:
starting timing test. nCns = 100, nSolvers = 10, nResolves = 50
done building data structures
time = 0.175
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.003
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.005
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.008
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.01
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.012
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.014
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.016
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.019
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.021
done adding 63 constraints [500 attempted, 0 exceptions]
time = 0.023
Editing vars with indices 70, 56
about to start resolves
done resolves -- now ending edits
total time = 0.325

  number of constraints:        100
  number of solvers:            10
  numbers of resolves:          50
  tests:                1
  time to add (ms):             23
  time to edit (ms):            5
  time to resolve (ms):         62
  time to edit (ms):            2
  add time per solver (ms):         0.023
  edit time per solver (ms):        0.25
  resolve time per resolve (ms):    0.124
  time to end edits per solver (ms):    0.1

[slightlyoff]

I found some more useless allocations/iteration. It's 8x faster now.

[d4tocchini]

that's significant!