Review Request 111643: RFC: Use QHash for TypeSystem registry

[Commit Hook]

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://git.reviewboard.kde.org/r/111643/
-----------------------------------------------------------

(Updated July 25, 2013, 7:30 p.m.)


Status
------

This change has been marked as submitted.


Review request for KDevelop.


Description
-------

Use a QHash instead of a fixed-size QVector for the type system registry.

While this is of course slower, the impact is imo negleglible. The benchmark
shows that the TypeSystem becomes ~20% slower, but that is still nothing in
total compared to other parts of KDevelop which are actually much slower.
Thus, limiting the API in such a drastic way (only up to N types) is imo
not worth it. Instead we can take the small performance impact and improve
the performance elsewhere :)

Also note that to encounter a slowdown of ~1s one would need to call the TypeSystem
functions approx. 45 million times. Setting a counter breakpoint on TypeSystem::create
shows that it's being called less than one million times when loading my biggest KDevelop
session. So yeah, I think we can safely ignore this time impact and classify the existing
code as a premature optimization.


Diffs
-----

  language/duchain/types/typeregister.h 4843eec 
  language/duchain/types/typeregister.cpp 90ed05f 

Diff: http://git.reviewboard.kde.org/r/111643/diff/


Testing
-------

Before this patch:

PASS   : TestDUChain::initTestCase()
RESULT : TestDUChain::benchTypeRegistry():"dataClassSize":
     0.0000299 msecs per iteration (total: 299, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"dynamicSize":
     0.0000438 msecs per iteration (total: 438, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"create":
     0.0001276 msecs per iteration (total: 1,276, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"isFactoryLoaded":
     0.0000271 msecs per iteration (total: 271, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"copy":
     0.0001812 msecs per iteration (total: 1,812, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"copyNonDynamic":
     0.0000409 msecs per iteration (total: 409, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"callDestructor":
     0.0000306 msecs per iteration (total: 306, iterations: 10000000)
PASS   : TestDUChain::benchTypeRegistry()
PASS   : TestDUChain::cleanupTestCase()
Totals: 3 passed, 0 failed, 0 skipped
********* Finished testing of TestDUChain *********

 Performance counter stats for './duchaintest benchTypeRegistry -iterations 10000000':

      10077.560833 task-clock                #    0.995 CPUs utilized          
               214 context-switches          #    0.021 K/sec                  
                36 cpu-migrations            #    0.004 K/sec                  
           171,631 page-faults               #    0.017 M/sec                  
    11,717,686,879 cycles                    #    1.163 GHz                     [83.35%]
     3,725,598,695 stalled-cycles-frontend   #   31.79% frontend cycles idle    [83.32%]
     1,212,888,069 stalled-cycles-backend    #   10.35% backend  cycles idle    [66.74%]
    19,470,323,749 instructions              #    1.66  insns per cycle        
                                             #    0.19  stalled cycles per insn [83.36%]
     5,693,920,745 branches                  #  565.010 M/sec                   [83.33%]
         1,995,259 branch-misses             #    0.04% of all branches         [83.27%]

      10.124920649 seconds time elapsed

After this patch:

PASS   : TestDUChain::initTestCase()
RESULT : TestDUChain::benchTypeRegistry():"dataClassSize":
     0.0000310 msecs per iteration (total: 310, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"dynamicSize":
     0.0000574 msecs per iteration (total: 574, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"create":
     0.0001497 msecs per iteration (total: 1,497, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"isFactoryLoaded":
     0.0000349 msecs per iteration (total: 349, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"copy":
     0.0001863 msecs per iteration (total: 1,863, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"copyNonDynamic":
     0.0000522 msecs per iteration (total: 522, iterations: 10000000)
RESULT : TestDUChain::benchTypeRegistry():"callDestructor":
     0.0000480 msecs per iteration (total: 480, iterations: 10000000)
PASS   : TestDUChain::benchTypeRegistry()
PASS   : TestDUChain::cleanupTestCase()
Totals: 3 passed, 0 failed, 0 skipped
********* Finished testing of TestDUChain *********

 Performance counter stats for './duchaintest benchTypeRegistry -iterations 10000000':

      11514.425401 task-clock                #    0.992 CPUs utilized          
               463 context-switches          #    0.040 K/sec                  
                65 cpu-migrations            #    0.006 K/sec                  
           171,639 page-faults               #    0.015 M/sec                  
    14,606,973,287 cycles                    #    1.269 GHz                     [83.36%]
     5,286,680,376 stalled-cycles-frontend   #   36.19% frontend cycles idle    [83.24%]
     2,277,868,635 stalled-cycles-backend    #   15.59% backend  cycles idle    [66.67%]
    21,485,714,055 instructions              #    1.47  insns per cycle        
                                             #    0.25  stalled cycles per insn [83.29%]
     6,111,305,415 branches                  #  530.752 M/sec                   [83.38%]
         2,157,996 branch-misses             #    0.04% of all branches         [83.35%]

      11.609370777 seconds time elapsed


Thanks,

Milian Wolff

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kdevelop-devel/attachments/20130725/fad21c87/attachment.html>