User:Stattrav/GSoC2012 log

28 May 2012

• The data is now pushed into the db. Takes the settings from a config file situated in the root of the benchmark source.
• Cleaning the script and testing with the data. I shall send in a formal mail discussing the script execution and usage which presents the work I've done.

27 May 2012

• Made the code completely based on the regexes as per the syntax in the bench/run.sh file.

24 May 2012

• Crude script for the parser and to transfer the data to the db.

22 May 2012

• Made progress with the parser.
• Code layout in terms of the folders was decided upon and is as follows:

 svnroot/brlcad/web/trunk/benchmark 
                                   |_ scripts
                                             |_ parser.py
                                   |_ sql
                                         |_ benchmark_db.sql

21 May 2012

• Completed the patch for the bench/pixcmp.c. Preliminary testing done. Benchmark script executes without fail. Waiting for Sean to approve and say if that is what is really wanted.

brlcad-build → bin/pixcmp -d4 ../brlcad/pix/moss.pix ../brlcad/pix/moss_new.pix
pixcmp pixels: 262140 matching, 4 off by 4, 0 off by many

brlcad-build → bin/pixcmp -d2 ../brlcad/pix/moss.pix ../brlcad/pix/moss.pix
pixcmp pixels: 262144 matching, 0 off by 2, 0 off by many

brlcad-build → bin/pixcmp -d2 ../brlcad/pix/moss.pix ../brlcad/pix/moss_new.pix
pixcmp pixels: 262140 matching, 4 off by 2, 0 off by many

brlcad-build → bin/pixcmp -d1 ../brlcad/pix/moss.pix ../brlcad/pix/moss_new.pix
pixcmp pixels: 262140 matching, 2 off by 1, 2 off by many

brlcad-build → bin/pixcmp ../brlcad/pix/moss.pix ../brlcad/pix/moss_new.pix
pixcmp pixels: 262140 matching, 2 off by 1, 2 off by many  

• Started writing the log parser.

18 May 2012

• Started looking into the issue which Sean mentioned which is related to pixcmp and found nothing there. Sean later found out that the bug was in the pixdiff utilty.
• Amidst this, I found that after the recent (svn) update of the code the benchmark test for sphflake was failing. I then spent time on figuring out(using binary search) the bug and found that r50582 was responsible for the same.

17 May 2012

• Working on a patch for the bench/pixcmp.c functionality.

15 May 2012

• Set up a dummy database without much changes to the schema mentioned as a part of the proposal and then started writing crude parser scripts in python.
• Setup thrift on my machine. Had faced few difficulties setting it up with php 5.4.3 and Python 3.2 The issues with php 5.4.3 were resolved but the ones with Python 3.2 were quite a few specially related to the syntactic changes during the transition from 2.x to 3.x. Downgraded python to 2.7.3 and completed the setup. Yet to test the interlanguage communication. Shall do it tomorrow.

14 May 2012

• Experimenting with regexes for the benchmark log parsing.

13 May 2012

• Moved from ubuntu 12.04 which I did not really like to Archlinux and setup the work environment.

12 May 2012

• Set up mediawiki on the local machine and then was experimenting with the api.php of mediawiki.

9 May 2012

• Continued experimentation with the benchmark script and finally understood what each part of the code does.
• Working on a back-end design (Sent in a mail to the developer mailing list). Shall start working on the db design.

8 May 2012

• Continued experimentation with benchmark script and attempting to create the first patch by fixing the following block in the run.sh script.

   # FIXME: should really iterate one file at a time so we don't
   # just zero-pad at the end
   perf_VGRREF=`grep RTFM $perf_ref_files | sed -n -e 's/^.*= *//' -e 's/ rays.*//p' | tr '\012' '\011' `
   perf_CURVALS=`grep RTFM $perf_cur_files | sed -n -e 's/^.*= *//' -e 's/ rays.*//p' | tr '\012' '\011' `

• Working on a set of questions for design discussions.

7 May 2012

• Still looking into benchmark script and experimenting with that.

4 May 2012

• Compiled the brlcad(r50443) on my new setup of Ubuntu 12.04. Installation went on smoothly when librtserver JDK support was disabled.(I was trying various options during compilation.) When turned on, there were the error of it not being able to find jni.h even though the paths were fine. Looking into it at the moment.
• Could not understand why cmake was looking for STDCXX libs.
• Ran the benchmark on the machine and sent the report to benchmark@brlcad.org
• Started looking into the run.sh script to figure out how it really works.