drd/scripts/run-splash2 - fp2-dev/platform/external/valgrind - Gitiles

 #!/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 > "$tmp"
   read avg1 stddev1 vsz1 vszdev1 < "$tmp"
   echo "Average time: ${avg1} +/- ${stddev1} seconds." \
        " VSZ: ${vsz1} +/- ${vszdev1} KB"

   p=2
   test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" | avgstddev > "$tmp"
   read avg2 stddev2 vsz2 vszdev2 < "$tmp"
   echo "Average time: ${avg2} +/- ${stddev2} seconds." \
        " VSZ: ${vsz2} +/- ${vszdev2} KB"

   p=4
   test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" | avgstddev > "$tmp"
   read avg4 stddev4 vsz4 vszdev4 < "$tmp"
   echo "Average time: ${avg4} +/- ${stddev4} seconds." \
        " VSZ: ${vsz4} +/- ${vszdev4} KB"

   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} \
     $VG --tool=drd --check-stack-var=yes "$@" -p${psep}${p} "${test_args}"

   p=4
   test_output="${1}-drd-without-stack-var-4.out" \
   print_runtime_ratio ${avg4} ${stddev4} ${vsz4} ${vszdev4} \
     $VG --tool=drd --check-stack-var=no  "$@" -p${psep}${p} "${test_args}"

   p=4
   test_output="${1}-helgrind-4.out" \
   print_runtime_ratio ${avg4} ${stddev4} ${vsz4} ${vszdev4} $VG --tool=helgrind "$@" -p${psep}${p} "${test_args}"

   echo ''

   rm -f "$tmp"
 }


 ########################
 # 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 -p4 versus -p4.
 # 17. Virtual memory size ratio for --tool=helgrind -p4 versus -p4.
 # 18. Execution time ratio for Intel Thread Checker -p4 versus -p4.
 # 19. 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
 ###########################################################################################################################
 # Results:                native       native       native       none      none       DRD       DRD      HG         ITC ITC
 #                         -p1          -p2          -p4           -p1       -p4       -p4      -p4+f     -p4      -p4 -p4+f
 # .........................................................................................................................
 # Cholesky                0.09  11880  0.07  21888  0.65  41883 15.2  5.08  3.9 2.15   17 2.41  13 2.60   50  5.86  239  82
 # FFT                     0.03   6560  0.02  15101  0.02  32176 10.7  8.36 17.0 2.48   92 2.88 113 3.13  156  5.43   90  41
 # LU, contiguous          0.07   3968  0.05  12176  0.06  28584 12.4 13.17 14.8 2.70   89 2.95  80 3.24  135  5.68  428 128
 # Ocean, contiguous       0.19  23208  0.18  31744  0.13  48888  4.8  3.45  7.6 2.17   59 2.58  50 2.75   77  4.01   90  28
 # Radix                   0.20  15008  0.11  23200  0.11  39592 13.1  4.22 23.6 2.23   59 2.67  54 2.87  282  6.23  222  56
 # Raytrace                0.66 206976  0.47 215168  0.47 232235  8.2  1.23 11.5 1.21  262 1.54 101 1.57  291  3.57  172  53
 # Water-n2                0.19  10560  0.09  26944  0.10  59704 11.4  5.65 22.0 1.83 3194 2.92 467 2.99  108  3.49  189  39
 # Water-sp                0.21   4312  0.10  13400  0.10  29496  9.8 12.20 20.7 2.63  254 3.14  64 3.42   99  5.13  183  34
 # .........................................................................................................................
 # geometric mean          0.41  13360  0.09  26319  0.12  48476 10.2  5.37 13.2 2.12  134 2.58  75 2.76  128  4.81  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, gcc 4.3.1, 32 bit executables, valgrind trunk r8795.
 ###########################################################################################################################

 ####
 # 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:
	#!/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 > "$tmp"
	read avg1 stddev1 vsz1 vszdev1 < "$tmp"
	echo "Average time: ${avg1} +/- ${stddev1} seconds." \
	" VSZ: ${vsz1} +/- ${vszdev1} KB"

	p=2
	test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" \| avgstddev > "$tmp"
	read avg2 stddev2 vsz2 vszdev2 < "$tmp"
	echo "Average time: ${avg2} +/- ${stddev2} seconds." \
	" VSZ: ${vsz2} +/- ${vszdev2} KB"

	p=4
	test_output="${1}-p${p}.out" measure_runtime "$@" -p${psep}${p} "${test_args}" \| avgstddev > "$tmp"
	read avg4 stddev4 vsz4 vszdev4 < "$tmp"
	echo "Average time: ${avg4} +/- ${stddev4} seconds." \
	" VSZ: ${vsz4} +/- ${vszdev4} KB"

	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} \
	$VG --tool=drd --check-stack-var=yes "$@" -p${psep}${p} "${test_args}"

	p=4
	test_output="${1}-drd-without-stack-var-4.out" \
	print_runtime_ratio ${avg4} ${stddev4} ${vsz4} ${vszdev4} \
	$VG --tool=drd --check-stack-var=no "$@" -p${psep}${p} "${test_args}"

	p=4
	test_output="${1}-helgrind-4.out" \
	print_runtime_ratio ${avg4} ${stddev4} ${vsz4} ${vszdev4} $VG --tool=helgrind "$@" -p${psep}${p} "${test_args}"

	echo ''

	rm -f "$tmp"
	}


	########################
	# 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 -p4 versus -p4.
	# 17. Virtual memory size ratio for --tool=helgrind -p4 versus -p4.
	# 18. Execution time ratio for Intel Thread Checker -p4 versus -p4.
	# 19. 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
	###########################################################################################################################
	# Results: native native native none none DRD DRD HG ITC ITC
	# -p1 -p2 -p4 -p1 -p4 -p4 -p4+f -p4 -p4 -p4+f
	# .........................................................................................................................
	# Cholesky 0.09 11880 0.07 21888 0.65 41883 15.2 5.08 3.9 2.15 17 2.41 13 2.60 50 5.86 239 82
	# FFT 0.03 6560 0.02 15101 0.02 32176 10.7 8.36 17.0 2.48 92 2.88 113 3.13 156 5.43 90 41
	# LU, contiguous 0.07 3968 0.05 12176 0.06 28584 12.4 13.17 14.8 2.70 89 2.95 80 3.24 135 5.68 428 128
	# Ocean, contiguous 0.19 23208 0.18 31744 0.13 48888 4.8 3.45 7.6 2.17 59 2.58 50 2.75 77 4.01 90 28
	# Radix 0.20 15008 0.11 23200 0.11 39592 13.1 4.22 23.6 2.23 59 2.67 54 2.87 282 6.23 222 56
	# Raytrace 0.66 206976 0.47 215168 0.47 232235 8.2 1.23 11.5 1.21 262 1.54 101 1.57 291 3.57 172 53
	# Water-n2 0.19 10560 0.09 26944 0.10 59704 11.4 5.65 22.0 1.83 3194 2.92 467 2.99 108 3.49 189 39
	# Water-sp 0.21 4312 0.10 13400 0.10 29496 9.8 12.20 20.7 2.63 254 3.14 64 3.42 99 5.13 183 34
	# .........................................................................................................................
	# geometric mean 0.41 13360 0.09 26319 0.12 48476 10.2 5.37 13.2 2.12 134 2.58 75 2.76 128 4.81 180 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, gcc 4.3.1, 32 bit executables, valgrind trunk r8795.
	###########################################################################################################################

	####
	# 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: