diff options
| -rw-r--r-- | README | 92 |
1 files changed, 92 insertions, 0 deletions
@@ -0,0 +1,92 @@ +Building +======== + +First you have to install the gsl and nlopt libraries. On Fedora you can +install these with the following command: + + $ yum install gsl gsl-devel NLopt NLopt-devel + +Then, to build everything you just type: + + $ make + +Some of the test programs also require the plot program which can be installed +with: + + $ yum install plotutils + +Fitting Events +============== + +You can fit events from SNOMAN data structure files by running the following +command: + + $ ./src/fit [input file] -o [output file] + +and then plot the fit results by using the plot script: + + $ ./utils/plot.py [output file] + +The plot script relies on the MC information in the SNOMAN file, so make sure +that SNOMAN is set up to output MC info when you simulate events (see the next +section). + +Simulating Events with SNOMAN +============================= + +Since the fitter reads in SNOMAN data structures it is necessary to use a +command file which correctly specifies the output file format. To benchmark the +fit results it is also necessary to save the initial MC vertex information. + +I have created a template command file in macros/mc_run_10000.cmd which +simulates run 10000 from the D2O phase. You can create a proper command file to +run SNOMAN with by running the snogen program which uses a simple templating +system to fill in some variables in this template. For example to simulate 100 +1 GeV muons uniformly distributed throughout the AV and with isotropic +directions you can run: + + $ ./macros/snogen -p mu_minus -e 1000.0 -n 100 | /path/to/snoman.exe + +See the comment at the top of ./macros/snogen for more information. + +Debugging +========= + +To debug memory issues you can compile everything with an address sanitizer. +Edit the Makefile and add the following flags: + + CFLAGS=-fsanitize=address + LDLIBS=-lasan + +and make sure that -lasan is the first library in the list. Then, recompile: + + $ make clean + $ make + +and run the fitter. It should crash if there is a buffer overflow or use after +free bug. + +You can also run the fitter with valgrind to search for memory leaks: + + $ valgrind --leak-check=full ./src/fit [input file] + +Profiling +========= + +To profile the code, you can use the linux perf command along with Brendan +Gregg's flamegraph software to produce a flame graph. + +First run the fitter: + + $ ./src/fit [input file] + +Then, identify the PID of the fit program and run: + + $ sudo perf record -F 99 -p [PID] --call-graph dwarf sleep 10 + +Then, to create the flame graph: + + $ sudo chown [username]:[username] perf.data + $ perf script > out.perf + $ stack-collapse.pl out.perf > out.folded + $ flamegraph.pl out.folded > kernel.svg |