Hi John, On Fri, Oct 23, 2020 at 11:46:54AM -0400, John Kacur wrote: > > TARGETS = $(sources:.c=) > > LIBS = -lrt -lpthread > > -RTTESTLIB = -lrttest -L$(OBJDIR) > > +RTTESTLIB = -lrttest -L$(OBJDIR) $(NUMA_LIBS) > > Currently only cyclictest was compiled with NUMA_LIBS, this change will > compile everything with NUMA_LIBS. I checked the size of the programs, and > they don't grow that much with this change, but they are small programs to > begin with, do we want to keep this functionality separate? My thinking is, that the most important program for testing seems to be cyclictest. Everyone will run cyclictest on the target platform. Thus libnuma will be available. So there wont be any new unresolved dependencies. I traded the size increase for simplification in the code base and build setup. Looking at the actual increase (x86_64, stripped) is not too bad: program old new diff --------------------------------------------------- cyclicdeadline 35488 35552 64 0.18% cyclictest 57632 57632 0 0.0% deadline_test 43712 43776 64 0.15% hackbench 19168 19168 0 0.0% oslat 36040 36072 32 0.089% pip_stress 27296 27360 64 0.23% pi_stress 44296 48456 4160 9.4% pmqtest 31864 31928 64 0.2% ptsematest 31752 31816 64 0.2% queuelat 14600 14600 0 0.0% rt-migrate-test 31696 31728 32 0.1% signaltest 31712 31776 64 0.2% sigwaittest 31792 31856 64 0.2% ssdd 14744 14744 0 0.0% svsematest 31856 31920 64 0.2% pi_stress is a bit odd though. Not sure what's happening there. Will look into it. So I would prefer to go this route and makes things simpler in the code base. Thanks, Daniel
