###############################################################################

.ONESHELL:

###############################################################################

ATT = $(EXE:=.att)
CAC = $(EXE:=.cac)
CAL = $(EXE:=.cal)
DAT = $(EXE:=.dat)
DEP = $(addprefix .,$(EXE:=.d))
EXE = $(basename $(SRC))
LOG = $(addsuffix .log,$(filter-out test,$(EXE)))
SRC = $(wildcard *.cc)
STA = $(addsuffix .sta,$(filter-out test,$(EXE)))
TXT = $(addsuffix .txt,$(filter-out test,$(EXE)))

###############################################################################

DEBUG = -fno-inline -Og
OPT = -O3
CXXFLAGS = -fanalyzer -g -MF .$*.d -MMD -pthread -std=c++20 -Wall $(OPT)
LDFLAGS = -L..
LDLIBS = -latomic -ldl -lrpmalloc

ifeq ($(findstring stm,$(notdir $(CURDIR))),stm)
	CXXFLAGS += -fgnu-tm
	DEBUG += -Os # trick to avoid undefined behavior!!!
else
	DEBUG += -fsanitize={address,alignment,bool,bounds,bounds-strict,builtin,enum,float-cast-overflow,float-divide-by-zero,integer-divide-by-zero,leak,nonnull-attribute,null,object-size,pointer-compare,pointer-overflow,pointer-subtract,return,returns-nonnull-attribute,shift,shift-base,shift-exponent,signed-integer-overflow,undefined,unreachable,vla-bound,vptr}
endif

# ifeq ($(findstring tag,$(notdir $(CURDIR))),tag)
# 	DEBUG = -fanalyzer -fno-inline -g -mno-{avx,sse} -Og # trick to avoid heap use after free!!!
# endif

###############################################################################

all: ld txt

att: $(ATT)

check: $(EXE)
	@for i in $^; do
		if nm -C ./$$i | grep -q rpmalloc; then
			echo "rpmalloc lib detected in $${PWD##*/}/$$i"
		else
			echo "rpmalloc lib NOT DETECTED in $${PWD##*/}/$$i!!!"
		fi
	done

clean:
	-rm -fv $(ATT) $(EXE) $(CAC) $(CAL) $(DAT) $(DEP) $(LOG) $(STA) $(TXT) callgrind.out.* core.* perf.* *.dat.old *~

dep: $(DEP)

exe: $(EXE)

ld:
	@LD_LIBRARY_PATH=$$LD_LIBRARY_PATH
	export LD_LIBRARY_PATH

log: $(LOG)

sta: $(STA)

txt: $(TXT)

###############################################################################

-include $(DEP)

%.att: %
	objdump -Cd $< > $@

%.cac: %
	valgrind --cachegrind-out-file=$@ --tool=cachegrind ./$<

%.cal: %
	valgrind --tool=callgrind --branch-sim=yes --cache-sim=yes --callgrind-out-file=$@ --collect-bus=yes --collect-jumps=yes --dump-instr=yes --dump-line=yes -- ./$<

%.cga: %.cac
	cg_annotate --auto=yes $<

%.dat: %
	perf record --call-graph dwarf --freq=max -o $@ -- ./$<

%.hel: %
	valgrind --tool=helgrind -- ./$<

%.kcg: %.cal
	kcachegrind $< &

%.leak: %
#	valgrind --leak-check=full -s --show-leak-kinds=all ./$<
	valgrind --leak-check=summary --show-leak-kinds=all ./$<

%.log %.sta: %
	@LANG=C perf stat -r 33 > $*.log 2> $*.sta -- ./$<

%.ltrace: %
	ltrace -cfS ./$< |& c++filt

%.perf: %.dat
	perf report -g 'graph,0.5,caller' -i $<

%.txt: %.log %.sta
	@v="[$$(cat $*.log | tr '\n' ',')]"
	sum=$$(python3 -c "import statistics; print('{0:.0f}'.format(sum($$v)))")
	median=$$(python3 -c "import statistics; print('{0:.0f}'.format(statistics.median($$v)))")
	mean=$$(python3 -c "import statistics; print('{0:.0f}'.format(statistics.mean($$v)))")
	stdev=$$(python3 -c "import statistics; print('{0:.0f}'.format(statistics.stdev($$v)))")
	cpu=$$(grep -o ".* msec task-clock" $*.sta | grep -o "[[:digit:].]*")
#	ratio=$$(python3 -c "print('{0:.2f}'.format($$sum / $$cpu))")
	ratio=$$(python3 -c "print('{0:.2f}'.format($$median / $$cpu))")
	hc=$$(grep 'N = 1;' stack.cc > /dev/null && echo '.' || echo ':')
	printf "%18s%10s%10s%10s%10s%10s\n" program median mean stdev cpu ratio | tee $@
	printf "%18s%10s%10s%10s%10s%10s\n" "$$(basename $$(pwd))$$hc" $$median $$mean $$stdev $$cpu $$ratio | tee -a $@

###############################################################################

.PHONY: all att check clean dep exe log sta txt
.PRECIOUS: $(CAC) $(CAL) $(DAT) $(LOG) $(STA) $(TXT)

###############################################################################