#!/bin/bash
########################
# Function definitions #
########################
source "$(dirname $0)/measurement-functions"
function run_test {
local tmp avg1 stddev1 avg2 stddev2 avg4 stddev4 p
tmp="/tmp/test-timing.$$"
rm -f "${tmp}"
p=1
test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" | avgstddev > "$tmp"
read avg1 stddev1 vsz1 vszdev1 rest < "$tmp"
echo "Average time: ${avg1} +/- ${stddev1} seconds." \
" VSZ: ${vsz1} +/- ${vszdev1} KB"
if [ "${rest}" != "" ]; then
echo "Internal error ($rest)"
exit 1
fi
rm -f "${tmp}"
p=2
test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" | avgstddev > "$tmp"
read avg2 stddev2 vsz2 vszdev2 rest < "$tmp"
echo "Average time: ${avg2} +/- ${stddev2} seconds." \
" VSZ: ${vsz2} +/- ${vszdev2} KB"
if [ "${rest}" != "" ]; then
echo "Internal error ($rest)"
exit 1
fi
rm -f "${tmp}"
p=4
test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" | avgstddev > "$tmp"
read avg4 stddev4 vsz4 vszdev4 rest < "$tmp"
echo "Average time: ${avg4} +/- ${stddev4} seconds." \
" VSZ: ${vsz4} +/- ${vszdev4} KB"
rm -f "$tmp"
if [ "${rest}" != "" ]; then
echo "Internal error ($rest)"
exit 1
fi
p=1
test_output="/dev/null" \
print_runtime_ratio "${avg1}" "${stddev1}" "${vsz1}" "${vszdev1}" "$VG" --tool=none "$@" -p${psep}${p} "${test_args}"
p=4
test_output="/dev/null" \
print_runtime_ratio "${avg4}" "${stddev4}" "${vsz4}" "${vszdev4}" "$VG" --tool=none "$@" -p${psep}${p} "${test_args}"
p=4
test_output="${1}-drd-with-stack-var-4.out" \
print_runtime_ratio "${avg4}" "${stddev4}" "${vsz4}" "${vszdev4}" \
"$VG" --tool=drd --first-race-only=yes --check-stack-var=yes \
--drd-stats=yes "$@" -p${psep}${p} "${test_args}"
p=4
test_output="${1}-drd-without-stack-var-4.out" \
print_runtime_ratio "${avg4}" "${stddev4}" "${vsz4}" "${vszdev4}" \
"$VG" --tool=drd --first-race-only=yes --check-stack-var=no \
--drd-stats=yes "$@" -p${psep}${p} "${test_args}"
p=4
test_output="${1}-helgrind-4-none.out" \
print_runtime_ratio "${avg4}" "${stddev4}" "${vsz4}" "${vszdev4}" "$VG" --tool=helgrind --history-level=none "$@" -p${psep}${p} "${test_args}"
p=4
test_output="${1}-helgrind-4-approx.out" \
print_runtime_ratio "${avg4}" "${stddev4}" "${vsz4}" "${vszdev4}" "$VG" --tool=helgrind --history-level=approx "$@" -p${psep}${p} "${test_args}"
p=4
test_output="${1}-helgrind-4-full.out" \
print_runtime_ratio "${avg4}" "${stddev4}" "${vsz4}" "${vszdev4}" "$VG" --tool=helgrind --history-level=full "$@" -p${psep}${p} "${test_args}"
echo ''
}
########################
# Script body #
########################
DRD_SCRIPTS_DIR="$(dirname $0)"
if [ "${DRD_SCRIPTS_DIR:0:1}" != "/" ]; then
DRD_SCRIPTS_DIR="$PWD/$DRD_SCRIPTS_DIR"
fi
SPLASH2="${DRD_SCRIPTS_DIR}/../splash2"
if [ ! -e "${SPLASH2}" ]; then
echo "Error: splash2 directory not found (${SPLASH2})."
exit 1
fi
if [ "$VG" = "" ]; then
VG="${DRD_SCRIPTS_DIR}/../../vg-in-place"
fi
if [ ! -e "$VG" ]; then
echo "Could not find $VG."
exit 1
fi
######################################################################################################################
# Meaning of the different colums:
# 1. SPLASH2 test name.
# 2. Execution time in seconds for native run with argument -p1.
# 3. Virtual memory size in KB for the native run with argument -p1.
# 4. Execution time in seconds for native run with argument -p2.
# 5. Virtual memory size in KB for the native run with argument -p2.
# 6. Execution time in seconds for native run with argument -p4.
# 7. Virtual memory size in KB for the native run with argument -p4.
# 8. Execution time ratio for --tool=none -p1 versus -p1.
# 9. Virtual memory size ratio for --tool=none -p1 versus -p1.
# 10. Execution time ratio for --tool=none -p4 versus -p4.
# 11. Virtual memory size ratio for --tool=none -p4 versus -p4.
# 12. Execution time ratio for --tool=drd --check-stack-var=yes -p4 versus -p4.
# 13. Virtual memory size ratio for --tool=drd --check-stack-var=yes -p4 versus -p4.
# 14. Execution time ratio for --tool=drd --check-stack-var=no -p4 versus -p4.
# 15. Virtual memory size ratio for --tool=drd --check-stack-var=no -p4 versus -p4.
# 16. Execution time ratio for --tool=helgrind --history-level=none -p4 versus -p4.
# 17. Virtual memory size ratio for --tool=helgrind --history-level=none -p4 versus -p4.
# 18. Execution time ratio for --tool=helgrind --history-level=approx -p4 versus -p4.
# 19. Virtual memory size ratio for --tool=helgrind --history-level=approx -p4 versus -p4.
# 20. Execution time ratio for --tool=helgrind --history-level=full -p4 versus -p4.
# 21. Virtual memory size ratio for --tool=helgrind --history-level=full -p4 versus -p4.
# 22. Execution time ratio for Intel Thread Checker -p4 versus -p4.
# 23. Execution time ratio for Intel Thread Checker -p4 versus -p4.
#
# Notes:
# - Both Helgrind and DRD use a granularity of one byte for data race detection.
# - Helgrind does detect data races on stack variables. DRD only detects
# data races on stack variables with --check-stack-var=yes.
# - The ITC tests have been run on a 4-way 2.5 GHz Pentium 4 workstation, most
# likely running a 32-bit OS. Not yet clear to me: which OS ? Which
# granularity does ITC use ? And which m4 macro's have been used by ITC as
# implementation of the synchronization primitives ?
#
# 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
################################################################################################################################################
# Results: native native native none none DRD DRD HG HG HG ITC ITC
# -p1 -p2 -p4 -p1 -p4 -p4 -p4+f -p4 -p4 -p4 -p4 -p4+f
# ..............................................................................................................................................
# Cholesky 0.11 12016 0.06 22016 0.55 41328 10.3 4.92 1.7 2.14 15 2.61 8 2.61 10 3.96 10 3.96 15 6.14 239 82
# FFT 0.02 6692 0.02 14888 0.02 31621 17.0 8.01 20.0 2.48 114 3.15 64 3.28 81 4.52 81 4.52 116 5.56 90 41
# LU, contiguous 0.08 4100 0.05 12304 0.06 28712 11.1 12.44 18.5 2.64 104 3.18 70 3.18 87 4.84 89 4.84 118 5.55 428 128
# Ocean, contiguous 0.23 16848 0.19 25384 0.23 42528 6.3 3.78 8.3 2.11 87 2.82 62 4.02 71 3.75 71 3.75 195 5.96 90 28
# Radix 0.21 15136 0.14 23336 0.15 39728 12.6 4.10 22.3 2.19 61 2.87 41 2.94 52 4.03 52 4.03 85 6.13 222 56
# Raytrace 0.63 207104 0.49 215296 0.49 231680 8.9 1.23 12.9 1.20 385 1.38 86 2.10 158 3.70 160 3.70 222 4.15 172 53
# Water-n2 0.18 10696 0.09 27072 0.11 59832 12.5 5.46 26.7 1.80 3092 3.03 263 3.06 92 3.28 92 3.28 92 3.55 189 39
# Water-sp 0.20 4444 0.15 13536 0.10 30269 10.6 11.56 27.0 2.52 405 3.29 69 3.42 95 4.59 95 4.59 97 4.73 183 34
# ..............................................................................................................................................
# geometric mean 0.14 13024 0.10 25669 0.14 47655 10.8 5.26 13.5 2.08 161 2.71 59 3.03 66 4.05 66 4.05 95 5.13 180 51
# ..............................................................................................................................................
# Hardware: dual-core Intel Core2 Duo E6750, 2.66 GHz, 4 MB L2 cache, 2 GB RAM.
# Software: openSUSE 11.0 (64-bit edition), runlevel 3, kernel 2.6.30.1, gcc 4.3.1, 32 bit SPLASH-2 executables, valgrind trunk r10648.
################################################################################################################################################
####
# Notes:
# - The ITC performance numbers in the above table originate from table 1 in
# the following paper:
# Paul Sack, Brian E. Bliss, Zhiqiang Ma, Paul Petersen, Josep Torrellas,
# Accurate and efficient filtering for the Intel thread checker race
# detector, Proceedings of the 1st workshop on Architectural and system
# support for improving software dependability, San Jose, California,
# 2006. Pages: 34 - 41.
# - The input parameters for benchmarks below originate from table 1 in the
# following paper:
# The SPLASH-2 programs: characterization and methodological considerations
# Woo, S.C.; Ohara, M.; Torrie, E.; Singh, J.P.; Gupta, A.
# 1995. Proceedings of the 22nd Annual International Symposium on Computer
# Architecture, 22-24 Jun 1995, Page(s): 24 - 36.
# ftp://www-flash.stanford.edu/pub/splash2/splash2_isca95.ps.Z
####
cache_size=$(get_cache_size)
log2_cache_size=$(log2 ${cache_size})
# Cholesky
(
cd ${SPLASH2}/codes/kernels/cholesky/inputs
for f in *Z
do
gzip -cd <$f >${f%.Z}
done
test_args=tk15.O run_test ../CHOLESKY -C$((cache_size))
)
# FFT
run_test ${SPLASH2}/codes/kernels/fft/FFT -t -l$((log2_cache_size/2)) -m16
# LU, contiguous blocks.
run_test ${SPLASH2}/codes/kernels/lu/contiguous_blocks/LU -n512
# LU, non-contiguous blocks.
#run_test ${SPLASH2}/codes/kernels/lu/non_contiguous_blocks/LU -n512
# Ocean
run_test ${SPLASH2}/codes/apps/ocean/contiguous_partitions/OCEAN -n258
#run_test ${SPLASH2}/codes/apps/ocean/non_contiguous_partitions/OCEAN -n258
# Radiosity. Runs fine on a 32-bit OS, but deadlocks on a 64-bit OS. Not clear to me why.
if [ $(uname -p) = "i686" ]; then
psep=' ' run_test ${SPLASH2}/codes/apps/radiosity/RADIOSITY -batch -room -ae 5000.0 -en 0.050 -bf 0.10
fi
# Radix
run_test ${SPLASH2}/codes/kernels/radix/RADIX -n$((2**20)) -r1024
# Raytrace
(
cd ${SPLASH2}/codes/apps/raytrace/inputs
rm -f *.env *.geo *.rl
for f in *Z
do
gzip -cd <$f >${f%.Z}
done
cd ..
test_args=inputs/car.env psep='' run_test ./RAYTRACE -m64
)
# Water-n2
(
cd ${SPLASH2}/codes/apps/water-nsquared
test_input=${DRD_SCRIPTS_DIR}/run-splash2-water-input psep=' ' run_test ./WATER-NSQUARED
)
# Water-sp
(
cd ${SPLASH2}/codes/apps/water-spatial
test_input=${DRD_SCRIPTS_DIR}/run-splash2-water-input psep=' ' run_test ./WATER-SPATIAL
)
# Local variables:
# compile-command: "./run-splash2"
# End: