add COMPILE_LTO to test link time optimization

the library performs best with lto since many small functions
can be inlined across objects.
This commit is contained in:
Daniel Mendler 2019-11-05 19:18:32 +01:00
parent c893d217aa
commit e8e65119d1
No known key found for this signature in database
GPG Key ID: D88ADB2A2693CA43
5 changed files with 20 additions and 10 deletions

View File

@ -163,6 +163,10 @@ matrix:
packages: packages:
- clang-4.0 - clang-4.0
# Link time optimization
- env: SANITIZER=1 COMPILE_LTO=1 BUILDOPTIONS='--with-cc=gcc-5 --with-m64 --with-travis-valgrind'
- env: SANITIZER=1 COMPILE_LTO=1 BUILDOPTIONS='--with-cc=clang-7 --with-m64 --with-travis-valgrind'
# GCC for the x86-64 architecture with restricted limb sizes # GCC for the x86-64 architecture with restricted limb sizes
# formerly started with the option "--with-low-mp" to testme.sh # formerly started with the option "--with-low-mp" to testme.sh
# but testing all three in one run took to long and timed out. # but testing all three in one run took to long and timed out.

View File

@ -3,7 +3,7 @@
void ndraw(const mp_int *a, const char *name) void ndraw(const mp_int *a, const char *name)
{ {
char *buf; char *buf;
size_t size; size_t size = 0;
mp_err err; mp_err err;
if ((err = mp_radix_size(a, 10, &size)) != MP_OKAY) { if ((err = mp_radix_size(a, 10, &size)) != MP_OKAY) {

View File

@ -77,7 +77,7 @@ profiled_single: pre_gen
rm -f *.o timing rm -f *.o timing
$(CC) $(LTM_CFLAGS) -fbranch-probabilities -c pre_gen/tommath_amalgam.c -o tommath_amalgam.o $(CC) $(LTM_CFLAGS) -fbranch-probabilities -c pre_gen/tommath_amalgam.c -o tommath_amalgam.o
$(AR) $(ARFLAGS) $(LIBNAME) tommath_amalgam.o $(AR) $(ARFLAGS) $(LIBNAME) tommath_amalgam.o
ranlib $(LIBNAME) $(RANLIB) $(LIBNAME)
install: $(LIBNAME) install: $(LIBNAME)
install -d $(DESTDIR)$(LIBPATH) install -d $(DESTDIR)$(LIBPATH)

View File

@ -86,6 +86,12 @@ LTM_CFLAGS += -O3 -funroll-loops
LTM_CFLAGS += -fomit-frame-pointer LTM_CFLAGS += -fomit-frame-pointer
endif endif
ifdef COMPILE_LTO
LTM_CFLAGS += -flto
AR = $(subst clang,llvm-ar,$(subst gcc,gcc-ar,$(CC)))
RANLIB = $(subst clang,llvm-ranlib,$(subst gcc,gcc-ranlib,$(CC)))
endif
endif # COMPILE_SIZE endif # COMPILE_SIZE
ifneq ($(findstring clang,$(CC)),) ifneq ($(findstring clang,$(CC)),)

View File

@ -107,7 +107,7 @@ _make()
{ {
echo -ne " Compile $1 $2" echo -ne " Compile $1 $2"
suffix=$(echo ${1}${2} | tr ' ' '_') suffix=$(echo ${1}${2} | tr ' ' '_')
CC="$1" CFLAGS="$2 $TEST_CFLAGS" make -j$MAKE_JOBS $3 $MAKE_OPTIONS > /dev/null 2>gcc_errors_${suffix}.log CC="$1" CFLAGS="$2 $TEST_CFLAGS" make -j$MAKE_JOBS $3 $MAKE_OPTIONS 2>gcc_errors_${suffix}.log
errcnt=$(wc -l < gcc_errors_${suffix}.log) errcnt=$(wc -l < gcc_errors_${suffix}.log)
if [[ ${errcnt} -gt 1 ]]; then if [[ ${errcnt} -gt 1 ]]; then
echo " failed" echo " failed"
@ -354,7 +354,7 @@ _banner
if [[ "$TEST_VS_MTEST" != "" ]] if [[ "$TEST_VS_MTEST" != "" ]]
then then
make clean > /dev/null make clean > /dev/null
_make "${compilers[0]} ${archflags[0]}" "$CFLAGS" "mtest_opponent" _make "${compilers[0]}" "${archflags[0]} $CFLAGS" "mtest_opponent"
echo echo
_make "gcc" "$MTEST_RAND" "mtest" _make "gcc" "$MTEST_RAND" "mtest"
echo echo
@ -394,15 +394,15 @@ do
fi fi
if [[ "$VALGRIND_BIN" != "" ]] if [[ "$VALGRIND_BIN" != "" ]]
then then
_runvalgrind "$i $a" "$CFLAGS" _runvalgrind "$i" "$a $CFLAGS"
[ "$WITH_LOW_MP" != "1" ] && continue [ "$WITH_LOW_MP" != "1" ] && continue
_runvalgrind "$i $a" "-DMP_16BIT $CFLAGS" _runvalgrind "$i" "$a -DMP_16BIT $CFLAGS"
_runvalgrind "$i $a" "-DMP_32BIT $CFLAGS" _runvalgrind "$i" "$a -DMP_32BIT $CFLAGS"
else else
_runtest "$i $a" "$CFLAGS" _runtest "$i" "$a $CFLAGS"
[ "$WITH_LOW_MP" != "1" ] && continue [ "$WITH_LOW_MP" != "1" ] && continue
_runtest "$i $a" "-DMP_16BIT $CFLAGS" _runtest "$i" "$a -DMP_16BIT $CFLAGS"
_runtest "$i $a" "-DMP_32BIT $CFLAGS" _runtest "$i" "$a -DMP_32BIT $CFLAGS"
fi fi
done done
done done