perfstat

#!/bin/sh

# measure performance difference with huge pages vs. without

let_cpu_cool()
{
	TEMP_LIMIT=$1
	echo "Waiting for CPU cores to cool to $TEMP_LIMIT degrees C"
	while (true)
	do
		sensors |\
		awk	'BEGIN { max = 0; }
			/^Core/ {
					if (0 + $3 > max) {
						max = 0 + $3;
					}
				}
			END {
				if (max > '"$1"')
					exit 1;
				else
					exit 0; }'
		if [ "$?" = "0" ]
		then
			break;
		fi
		sleep 1
	done
	echo "CPU temp has cooled to $TEMP_LIMIT degrees C"
}

run_test()
{
	CACHE_AWARE="$1"
	OUTPUTFILE="$2"
	echo "Running with $CACHE_AWARE fraction of cache-aware particles..."
	echo "START: $(date)"
	# perf stat invocation cribbed from https://rigtorp.se/hugepages/
	sudo perf stat -e 'faults,dTLB-loads,dTLB-load-misses,cache-misses,cache-references' \
		   ./gaseous-giganticus --noise-scale 2.5 --velocity-factor 1300 --bands 10 \
			-i input_image.png -o output_image -c 300 -K $CACHE_AWARE > "$OUTPUTFILE" 2>&1
	echo "END: $(date)"
}

let_cpu_cool 35
run_test 0.0 test01.txt
let_cpu_cool 35
run_test 0.5 test02.txt
let_cpu_cool 35
run_test 1.0 test03.txt

grep misses test0[123].txt