mirror of
https://github.com/cuberite/libdeflate.git
synced 2025-09-09 12:16:42 -04:00
tools/checksum_benchmarks.sh: various improvements
Make it compatible with the new code organization, make it run the test_checksums program for each implementation, and run each implementation in both 64-bit and 32-bit modes.
This commit is contained in:
Eric Biggers
parent
bf0797e666
commit
e7aa4666e0
@ -4,6 +4,9 @@ set -eu
|
||||
|
||||
if [ $# -eq 0 ]; then
|
||||
FILE="$HOME/data/silesia"
|
||||
[ ! -e "$FILE" ] && FILE="$HOME/silesia"
|
||||
[ ! -e "$FILE" ] && FILE="$HOME/data/testdata"
|
||||
[ ! -e "$FILE" ] && FILE="$HOME/testdata"
|
||||
echo "Using default FILE: $FILE"
|
||||
echo
|
||||
elif [ $# -eq 1 ]; then
|
||||
@ -13,64 +16,126 @@ else
|
||||
exit 1
|
||||
fi
|
||||
|
||||
ARCH="$(uname -m)"
|
||||
|
||||
if ! grep -q '\<sse2\>' /proc/cpuinfo; then
|
||||
echo "This script must be run on an x86 CPU" 1>&2
|
||||
if [ -n "$(git status -s | egrep 'adler32_impl.h|crc32_impl.h')" ]; then
|
||||
echo "This script will overwrite adler32_impl.h and crc32_impl.h," \
|
||||
"which have uncommitted changes. Refusing to run." 1>&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
if [ -n "$(git status -s | egrep 'adler32.c|crc32.c')" ]; then
|
||||
echo "This script will overwrite adler32.c and crc32.c, which" \
|
||||
"have uncommitted changes. Refusing to run." 1>&2
|
||||
exit 1
|
||||
fi
|
||||
have_cpu_feature() {
|
||||
local feature="$1"
|
||||
local tag
|
||||
case $ARCH in
|
||||
arm*|aarch*)
|
||||
tag="Features"
|
||||
;;
|
||||
*)
|
||||
tag="flags"
|
||||
;;
|
||||
esac
|
||||
grep -q "^$tag"$'[ \t]'"*:.*\<$feature\>" /proc/cpuinfo
|
||||
}
|
||||
|
||||
make_and_test() {
|
||||
make "$@" checksum test_checksums > /dev/null
|
||||
./test_checksums > /dev/null
|
||||
}
|
||||
|
||||
__do_benchmark() {
|
||||
local impl="$1" speed
|
||||
shift
|
||||
local flags="$CKSUM_FLAGS $*"
|
||||
|
||||
speed=$(./checksum $flags -t "$FILE" | \
|
||||
grep -o '[0-9]\+ MB/s' | grep -o '[0-9]\+')
|
||||
printf "%-45s%-10s\n" "$CKSUM_NAME ($impl)" "$speed"
|
||||
}
|
||||
|
||||
do_benchmark() {
|
||||
method="$1"
|
||||
shift
|
||||
speed=$(./checksum "$@" -t "$FILE" | \
|
||||
grep -o '[0-9]\+ MB/s' | grep -o '[0-9]\+')
|
||||
printf "%-36s%-10s\n" "$method" "$speed"
|
||||
local impl="$1"
|
||||
|
||||
if [ "$impl" = zlib ]; then
|
||||
__do_benchmark "$impl" "-Z"
|
||||
else
|
||||
make_and_test CFLAGS="$EXTRA_CFLAGS"
|
||||
__do_benchmark "libdeflate, $impl"
|
||||
if [ $ARCH = x86_64 ]; then
|
||||
make_and_test CFLAGS="-m32 $EXTRA_CFLAGS"
|
||||
__do_benchmark "libdeflate, $impl, 32-bit"
|
||||
fi
|
||||
fi
|
||||
}
|
||||
|
||||
sort_by_speed() {
|
||||
awk '{print $NF, $0}' | sort -nr | cut -f2- -d' '
|
||||
}
|
||||
|
||||
disable_impl() {
|
||||
local name="$1"
|
||||
local extra_cflags="$2"
|
||||
|
||||
sed -i '/^\#ifdef DISPATCH_'"$name"'$/aif (0)' lib/*/{adler,crc}32_impl.h
|
||||
EXTRA_CFLAGS+=" $extra_cflags"
|
||||
}
|
||||
|
||||
restore_impls() {
|
||||
git checkout -f lib/*/{adler,crc}32_impl.h
|
||||
}
|
||||
|
||||
trap restore_impls EXIT
|
||||
|
||||
cat << EOF
|
||||
Method Speed (MB/s)
|
||||
------ ------------
|
||||
Method Speed (MB/s)
|
||||
------ ------------
|
||||
EOF
|
||||
|
||||
# CRC-32
|
||||
(
|
||||
if grep -q '\<pclmulqdq\>' /proc/cpuinfo; then
|
||||
make checksum > /dev/null
|
||||
do_benchmark "CRC-32 (libdeflate, PCLMUL/AVX)"
|
||||
sed -i '/^\#if NEED_PCLMUL_IMPL && !defined(__AVX__)/,/^\#endif$/d' lib/crc32.c
|
||||
make checksum > /dev/null
|
||||
do_benchmark "CRC-32 (libdeflate, PCLMUL)"
|
||||
fi
|
||||
sed -i '/^#if defined(__PCLMUL__)/,/^\#endif$/d' lib/crc32.c
|
||||
make checksum > /dev/null
|
||||
do_benchmark "CRC-32 (libdeflate, generic)"
|
||||
git checkout -f lib/crc32.c > /dev/null
|
||||
do_benchmark "CRC-32 (zlib)" -Z
|
||||
) | sort_by_speed
|
||||
CKSUM_NAME="CRC-32"
|
||||
CKSUM_FLAGS=""
|
||||
EXTRA_CFLAGS=""
|
||||
{
|
||||
case $ARCH in
|
||||
i386|x86_64)
|
||||
if have_cpu_feature pclmulqdq && have_cpu_feature avx; then
|
||||
do_benchmark "PCLMUL/AVX"
|
||||
disable_impl "PCLMUL_AVX" "-mno-avx"
|
||||
fi
|
||||
if have_cpu_feature pclmulqdq; then
|
||||
do_benchmark "PCLMUL"
|
||||
disable_impl "PCLMUL" "-mno-pclmul"
|
||||
fi
|
||||
;;
|
||||
esac
|
||||
do_benchmark "generic"
|
||||
do_benchmark "zlib"
|
||||
} | sort_by_speed
|
||||
|
||||
# Adler-32
|
||||
CKSUM_NAME="Adler-32"
|
||||
CKSUM_FLAGS="-A"
|
||||
EXTRA_CFLAGS=""
|
||||
echo
|
||||
(
|
||||
if grep -q '\<avx2\>' /proc/cpuinfo; then
|
||||
make checksum > /dev/null
|
||||
do_benchmark "Adler-32 (libdeflate, AVX2)" -A
|
||||
fi
|
||||
sed -i '/^#if defined(__AVX2__)/,/^\#endif$/d' lib/adler32.c
|
||||
make checksum > /dev/null
|
||||
do_benchmark "Adler-32 (libdeflate, SSE2)" -A
|
||||
sed -i '/^#ifdef __SSE2__/,/^\#endif$/d' lib/adler32.c
|
||||
make checksum > /dev/null
|
||||
do_benchmark "Adler-32 (libdeflate, generic)" -A
|
||||
git checkout -f lib/adler32.c > /dev/null
|
||||
do_benchmark "Adler-32 (zlib)" -A -Z
|
||||
) | sort_by_speed
|
||||
{
|
||||
case $ARCH in
|
||||
i386|x86_64)
|
||||
if have_cpu_feature avx2; then
|
||||
do_benchmark "AVX2"
|
||||
disable_impl "AVX2" "-mno-avx2"
|
||||
fi
|
||||
if have_cpu_feature sse2; then
|
||||
do_benchmark "SSE2"
|
||||
disable_impl "SSE2" "-mno-sse2"
|
||||
fi
|
||||
;;
|
||||
arm*|aarch*)
|
||||
if have_cpu_feature neon; then
|
||||
do_benchmark "NEON"
|
||||
disable_impl "NEON" "-mfpu=vfpv3"
|
||||
fi
|
||||
;;
|
||||
esac
|
||||
do_benchmark "generic"
|
||||
do_benchmark "zlib"
|
||||
} | sort_by_speed
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user