# This is a SuperMongo macro file. To run these macros make sure you have
# majurics smtools installed.
#
# Call the macro 'benchmark' for automated benchmarking. The resulting
# figures will be stored in figs/ directory, with the datafiles in the data/
# directory.
#

rr
	re macros.sm

benchmark
	! (cd .. && make)
	! mkdir -p figs data
	init
	benchcalc
	benchsearch
	mkfigs

init
	! ./getdevname.sh > deviceid.txt
	data deviceid.txt
	read '%s %s %30s' <devnameu host devname>
	define GPU  (devname[0])
	define GPUU (devnameu[0])
	define HOST (host[0])
	! rm -f deviceid.txt

benchcalc
	! cat ../wordlist.txt | ../cuda_md5 --benchmark | sed -r 's/ *[^ ]+=/\t/g' > data/bench.$HOST.calc.dat

benchsearch
	! cat ../wordlist.txt | ../cuda_md5 --benchmark --search=AliceBob | sed -r 's/ *[^ ]+=/\t/g' > data/bench.$HOST.search.dat

tpbplot 1
	# Plot of computation speed vs. number of threads per block
	# See README for instructions on how to produce bench.dat
	#
	data "data/bench.$!HOST.$!1.dat"
	read <tpb 2 nblocks 4 padding 7 dynmem 8 shmem 9 t 10 v 11>
	set bpmp = int(2**14 / shmem)

	lweight 3
	expand 1.0001
	limits tpb v
	erase box 4 2 0 4
	connect tpb v

	limits tpb 0 10
	box 1 0 0 2
	connect tpb bpmp

	ltype 1	
	do i=0,256,32 { vline $i\n }
	ltype 0

	xlabel Threads per block
	ylabel \raise500MD5 hashes/second
	relocate $fx2 $(0.5*($fy2+$fy1))
	angle 90
	putlabel 2 \raise-1500Blocks per multiprocessor
	angle 0

	xuplabel "$!HOST : $!GPU ($!1)"
	lweight 1

mkfigs
	psfile figs/perf_vs_tpb.$GPUU.$HOST.calc.ps tpbplot calc
	psfile figs/perf_vs_tpb.$GPUU.$HOST.search.ps tpbplot search

	! convert figs/perf_vs_tpb.$GPUU.$HOST.calc.ps figs/perf_vs_tpb.$GPUU.$HOST.calc.png
	! convert figs/perf_vs_tpb.$GPUU.$HOST.search.ps figs/perf_vs_tpb.$GPUU.$HOST.search.png