tensorflow_blt

#!/bin/sh

set -eu

trap exit TERM INT QUIT
trap "killall tensorflow_mnist_short 2>/dev/null || true" EXIT

show_usage="0"
if [ ! -x setpgrp ]
then
	echo "ERROR: setpgrp binary not found; did you build it?" >&2
	exit 3
fi
if [ -z "${CATBENCH_JAGFILE-}" ]
then
	cat >&2 <<-tac
		$0: should be invoked via driver

	tac
	show_usage="2"
fi
if [ "$show_usage" -ne "0" -o $# -gt 2 ]
then
	cat <<-tac
		USAGE: $0 [min core count] [max core count]

		[min core count] defaults to one.
		[max core count] defaults to one fewer than the number of physical cores on the first CPU socket.
		If only one of these is specified, it is assumed to be the max.
	tac
	exit "$show_usage"
fi

logfilename="$CATBENCH_JAGFILE"
num_trash="`lscpu | grep Core | tr -s " " | rev | cut -d" " -f1 | rev`"
step_trash="`grep -m2 "physical id" /proc/cpuinfo | uniq | wc -l`"
start_trash="1"
if [ -n "${2-}" ]
then
	start_trash="$1"
	shift
fi
if [ -n "${1-}" ]
then
	num_trash="$1"
fi
constrain="$(((1 << (step_trash * num_trash)) - 1))"
printf "%s\n" "Trying core counts between $start_trash and $num_trash..."

jaguar/jaguar set "$logfilename" legend.samples.num_trash.description string "Cores running contender"
jaguar/jaguar set "$logfilename" legend.samples.num_trash.unit string ""

jaguar/jaguar set "$logfilename" legend.samples.contender_tput.description string "Contender throughput"
jaguar/jaguar set "$logfilename" legend.samples.contender_tput.unit string "epochs/s"

jaguar/jaguar set "$logfilename" data.basecontender.description string "ContenderOnly"

data="integer,fraction\nnum_trash,contender_tput"
mask="0x0"
for num in `seq "$num_trash"`
do
	mask="`printf %#x "$(((mask + ((1 << (step_trash * num)) | num == num_trash)) & constrain))"`"
	if [ "$num" -lt "$start_trash" ]
	then
		continue
	fi

	printf "\n%s\n" "$num cores (mask $mask):"
	time="`time -p taskset "$mask" ./setpgrp ./tensorflow_mnist_short 2>&1 | tee /dev/tty | sed -n 's/.*real \(.*\).*/\1/p'`"
	tput="`printf "%s\n" "1 / $time" | bc -l`"
	printf "%s\n" "time $time s (tput $tput epochs/s)"
	data="$data\n$num,$tput"
done

printf "$data" | jaguar/jaguar set "$logfilename" data.basecontender.samples array -