Here is the latest set of numbers on the GHC repo.  I left out the check
test because it takes too long.

Note: in my last table, I called the --disable-bytestring version of
pre-salvo-8 "old"; what I should have called it was "fps" instead, which
I have done here.

The only real difference we should see in this table is in the
pull-1000 test, which now really manipulates 1000 patches.  The
speedup seems quite dramatic.  Is that an error, or is life actually
good?


                  || fps    | new
==================++========+======
     get-complete ||    7.5 |   5.7
             init ||    0.0 |   0.0
        add-r-ghc ||    0.4 |   0.3
         get-lazy ||    0.5 |   0.5
        unpull-10 ||    9.1 |   7.4
          pull-10 ||    3.4 |   2.6
       unpull-100 ||   62.6 |  15.0
         pull-100 ||   72.3 |  27.3
      unpull-1000 ||  211.3 |  28.8
        pull-1000 ||  189.9 |  49.4
unpull-1000-cache ||   61.3 |  11.5
  pull-1000-cache ||  190.6 |  49.3
   whatsnew-empty ||    0.1 |   0.1
  whatsnew-id-ego ||    1.8 |   1.7
           revert ||    7.6 |   5.7
           record ||    5.2 |   4.6
         unrecord ||   11.4 |   5.5
         record-l ||    5.3 |   4.3
         annotate ||  985.7 | 755.2

Next, I will be adding the darcs darcs repository to the set of Standard
Darcs Benchmarks so that we can more easily reproduce the annotate
Setup.hs regression.

I don't really understand the difference between the two unpull
tests by the way, since I wouldn't expect the cache to actually
impact anything...  I'll attach also the shell script on which this test is 
based.

-- 
Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow>
PGP Key ID: 08AC04F9

Attachment: go-one-iteration.sh
Description: Bourne shell script

Attachment: signature.asc
Description: Digital signature

_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users

Reply via email to