GitLab

Implements copy-on-write as discussed in #5 (closed). This is still a work-in-progress.

The test test_BinaryMethod seems to fail from time to time. I have to further investigate this.

      Start 19: Scenario: Calls to functions of super class

19: Test command: /home/jp/Programming/MA/elsa/build/elsa/projectors/tests/test_BinaryMethod "Scenario: Calls to functions of super class"
19: Test timeout computed to be: 9.99988e+06
19: Filters: Scenario: Calls to functions of super class
19: [2019-10-09 17:18:21.188] [Timing] [info] Execution of BinaryMethod: apply took 4 ms
19: [2019-10-09 17:18:21.204] [Timing] [info] Execution of BinaryMethod: apply took 15 ms
19:
19: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
19: test_BinaryMethod is a Catch v2.8.0 host application.
19: Run with -? for options
19:
19: -------------------------------------------------------------------------------
19: Scenario: Calls to functions of super class
19:       Given: A projector
19:        When: Projector is cloned
19:        Then: Results do not change (may still be slightly different due to
19:              summation being performed in a different order)
19: -------------------------------------------------------------------------------
19: /home/jp/Programming/MA/elsa/elsa/projectors/tests/test_BinaryMethod.cpp:465
19: ...............................................................................
19:
19: /home/jp/Programming/MA/elsa/elsa/projectors/tests/test_BinaryMethod.cpp:468: FAILED:
19:   REQUIRE( sino == sinoClone )
19: with expansion:
19:   {?} == {?}
19:
19: ===============================================================================
19: test cases: 1 | 1 failed
19: assertions: 1 | 1 failed
19:
19/36 Test #19: Scenario: Calls to functions of super class ..................................***Failed    0.02 sec

Benchmark results

I did a quick memory profiling with valgrind --tool=massif ... using the example2d. It seems that the CoW implementation does its job. Further benchmarking with larger examples necessary.

Results with CoW:

--------------------------------------------------------------------------------
  n        time(i)         total(B)   useful-heap(B) extra-heap(B)    stacks(B)
--------------------------------------------------------------------------------
 65    637,923,233          514,224          450,874        63,350            0

Results w/o CoW:

--------------------------------------------------------------------------------
  n        time(i)         total(B)   useful-heap(B) extra-heap(B)    stacks(B)
--------------------------------------------------------------------------------
 81    568,530,524          784,632          721,218        63,414            0

Next steps

• investigate test failure
• write test cases
• extensive benchmarks