Add -O3 flag when compiling crypto libraries.

The optimized code is surprisingly fast, only a small percent slower
than specialized instructions, which can't really be used with an
amalgamation build of Botan.

This makes DS2 regression tests way faster.

oscar/oscar.pro

@@ -311,8 +311,6 @@ SOURCES += \
     SleepLib/loader_plugins/somnopose_loader.cpp \
     SleepLib/loader_plugins/viatom_loader.cpp \
     SleepLib/loader_plugins/zeo_loader.cpp \
-    SleepLib/thirdparty/botan_all.cpp \
-    SleepLib/crypto.cpp \
     zip.cpp \
     SleepLib/thirdparty/miniz.c \
     csv.cpp \
@@ -343,6 +341,19 @@ SOURCES += \
     SOURCES += help.cpp
 }
 
+# The crypto libraries need to be optimized to avoid a 5x slowdown or worse.
+# Don't use this for everything, as it interferes with debugging.
+SOURCES_OPTIMIZE = \
+    SleepLib/thirdparty/botan_all.cpp \
+    SleepLib/crypto.cpp
+optimize.name = optimize
+optimize.input = SOURCES_OPTIMIZE
+optimize.dependency_type = TYPE_C
+optimize.variable_out = OBJECTS
+optimize.output = ${QMAKE_VAR_OBJECTS_DIR}${QMAKE_FILE_IN_BASE}$${first(QMAKE_EXT_OBJ)}
+optimize.commands = $${QMAKE_CXX} -c $(CXXFLAGS) -O3 $(INCPATH) -o ${QMAKE_FILE_OUT} ${QMAKE_FILE_IN}
+QMAKE_EXTRA_COMPILERS += optimize
+
 HEADERS  += \
     checkupdates.h \
     common_gui.h \