diff --git a/Makefile b/Makefile
index 2c85d576ab..09eb1add17 100644
--- a/Makefile
+++ b/Makefile
@@ -20,6 +20,7 @@
 # * -DUPB_UNALIGNED_READS_OK: makes code smaller, but not standard compliant
 
 .PHONY: all lib clean tests test benchmarks benchmark descriptorgen
+.PHONY: clean_leave_profile
 
 # Default rule: just build libupb.
 all: lib
@@ -110,9 +111,9 @@ ALLSRC=$(CORE) $(STREAM) $(BENCHMARKS_SRC) $(TESTS_SRC)
 
 # Rules. #######################################################################
 
-clean:
+clean_leave_profile:
 	rm -rf $(LIBUPB) $(LIBUPB_PIC)
-	rm -rf $(call rwildcard,,*.o) $(call rwildcard,,*.lo) $(call rwildcard,,*.gcno) $(call rwildcard,,*.dSYM)
+	rm -rf $(call rwildcard,,*.o) $(call rwildcard,,*.lo) $(call rwildcard,,*.dSYM)
 	rm -rf benchmark/google_messages.proto.pb benchmark/google_messages.pb.* benchmarks/b.* benchmarks/*.pb*
 	rm -rf $(TESTS) tests/t.*
 	rm -rf src/descriptor.pb
@@ -120,6 +121,9 @@ clean:
 	rm -rf lang_ext/lua/upb.so
 	cd lang_ext/python && python setup.py clean --all
 
+clean: clean_leave_profile
+	rm -rf $(call rwildcard,,*.gcno) $(call rwildcard,,*.gcda)
+
 # Core library (libupb.a).
 SRC=$(CORE) $(STREAM)
 LIBUPB=src/libupb.a
diff --git a/perf-tests.sh b/perf-tests.sh
index 003f866e61..8630182fbb 100755
--- a/perf-tests.sh
+++ b/perf-tests.sh
@@ -11,24 +11,25 @@ fi
 
 rm -f perf-tests.out
 
-if [ x`uname -m` = xx86_64 ]; then
+run_with_flags () {
+  FLAGS=$1
+  NAME=$2
+
   make clean
-  echo "-DNDEBUG -m32" > perf-cppflags
+  echo "$FLAGS -fprofile-generate" > perf-cppflags
   make upb_benchmarks
-  make benchmark | sed -e 's/^/plain32./g' | tee -a perf-tests.out
+  make benchmark
 
-  make clean
-  echo "-DNDEBUG -fomit-frame-pointer -m32" > perf-cppflags
+  make clean_leave_profile
+  echo "$FLAGS -fprofile-use" > perf-cppflags
   make upb_benchmarks
-  make benchmark | sed -e 's/^/omitfp32./g' | tee -a perf-tests.out
-fi
+  make benchmark | sed -e "s/^/$NAME./g" | tee -a perf-tests.out
+}
 
-make clean
-echo "-DNDEBUG" > perf-cppflags
-make $MAKETARGET
-make benchmark | sed -e 's/^/plain./g' | tee -a perf-tests.out
+if [ x`uname -m` = xx86_64 ]; then
+  run_with_flags "-DNDEBUG -m32" "plain32"
+  run_with_flags "-DNDEBUG -fomit-frame-pointer -m32" "omitfp32"
+fi
 
-make clean
-echo "-DNDEBUG -fomit-frame-pointer" > perf-cppflags
-make $MAKETARGET
-make benchmark | sed -e 's/^/omitfp./g' | tee -a perf-tests.out
+run_with_flags "-DNDEBUG " "plain"
+run_with_flags "-DNDEBUG -fomit-frame-pointer" "omitfp"