#!/bin/bash

rm -f bench.bin core core.*

if [ "$1" = "-new" -o "$1" = "-clean" ] ; then
	rm -f cache.txt
	[ "$1" = "-clean" ] && exit 0
elif [ "$1" != "-resume" ] ; then
	echo "Usage: $0 { -new | -resume | -clean } [ compiler ]"
	exit 1
fi

compiler=${2:-gcc}

declare -a options=()
did_new_tests=1

tryit() {
	thisval=$( egrep "^[0-9]+: $*\$" cache.txt 2> /dev/null | cut -f1 -d: )
	[ -z "$thisval" ] || return

	tmp=`mktemp`
	echo -e "\nTesting: $* ..."
	if ! $* *.c *.cc bzlib/*.c -lstdc++ -lm -Ibzlib -o bench.bin &> $tmp || [ -s $tmp ]
	then
		cat $tmp ; rm $tmp ; thisval=0
		echo "Compiler returned an error!"
		echo "0: $*" >> cache.txt
		return
	fi
	if ! ./bench.bin > $tmp || [ ! -s $tmp ]
	then
		rm $tmp ; thisval=0
		echo "Benchmark returned an error!"
		echo "0: $*" >> cache.txt
		return
	fi
	thisval=$( cat $tmp )
	echo "$thisval: $*" >> cache.txt
	did_new_tests=1
	rm $tmp
}

try() {
	bestval=0 ; best=""
	for x in "" "$@" ; do
		options[options_counter]="$x"
		tryit $compiler ${options[*]}
		if [ $thisval -gt $bestval ] ; then
			bestval=$thisval ; best="$x"
		fi
	done
	options[options_counter]="$best"
	(( options_counter++ ))
}

while
	[ $did_new_tests -eq 1 ]
do
	options_counter=0
	did_new_tests=0

	try -march={i386,i486,i586,i686,pentium{-mmx,2,3,4}} \
	    -march={k6,k6-2,k6-3,athlon,athlon-{tbird,4,xp,mp}}

	# Options from -O1
	#
	try -fmerge-constants
	try -fdefer-pop
	try -fthread-jumps
	try -fdelayed-branch
	try -fomit-frame-pointer
	try -fguess-branch-prob
	try -fcprop-registers


	# Options from -O2
	#
	# Note: -fschedule-insns2 is not used in -O2
	#
	try -foptimize-sibling-calls
	try -fcse-follow-jumps
	try -fcse-skip-blocks
	try -fgcse
	try -fexpensive-optimizations
	try -fstrength-reduce
	try -frerun-cse-after-loop
	try -frerun-loop-opt
	try -fcaller-saves
	try -fforce-mem
	try -fpeephole2
	try -fschedule-insns -fschedule-insns2
	try -fregmove
	try -fstrict-aliasing
	try -fdelete-null-pointer-checks
	try -freorder-blocks

	# Options from -O3
	#
	try -finline-functions
	try -frename-registers

	# Options from -Os
	#
	try -fmerge-constants
	try -falign-loops
	try -falign-jumps
	try -falign-labels
	try -falign-functions

	# Other nice options
	#
	try -fforce-addr
	try -funroll-loops -funroll-all-loops
	try -maccumulate-outgoing-args -mno-push-args
	try -preferred-stack-boundary={2,3,4}
	try -fprefetch-loop-arrays
	try -fmove-all-movables
	try -freduce-all-givs

	# Dangerous options which are not included by default
	#
	# try -malign-double

	# Options from -ffast-math
	#
	# try -funsafe-math-optimizations
	# try -fno-math-errno
	# try -fno-trapping-math

	# Other floating point math options
	#
	# try -mfpmath=387
	# try -mfpmath=sse
	# try -mfpmath=sse,387
	# try -mieee-fp
done

rm -f bench.bin core core.*
echo -e "\nCompiler benchmark finished. Best results with:"
echo "$compiler" ${options[*]} ; echo