Tell gcc to avoid floating point contraction

On ppc64el Ubuntu, gcc was using fmadd instead of fmul which caused
'-459.67 °F' to be parsed as '-8.18283712394279 fK' instead of plain
'0 K'. The same problem with Reaumur could be fixed by adding extra
precision digits to the numbers in the temperature.sql file, but that
didn't work for °F. Turn add/mul contraction off for the nicer-looking
result.

Author: df7cb

Makefile

@@ -8,6 +8,10 @@ DATA_built = unit--2--3.sql unit--3.sql
 REGRESS = extension tables unit prefix units temperature functions derived compare aggregate iec custom
 EXTRA_CLEAN = unitparse.yy.* powers powers.o unit-*.dump # unitparse.tab.*
 
+# avoid add/mult contraction so '-459.67 °F' is really '0 K'
+# problem visible on ppc64el Ubuntu trusty..zesty (Debian unaffected)
+PG_CPPFLAGS += -ffp-contract=off
+
 PG_CONFIG = pg_config
 PGXS := $(shell $(PG_CONFIG) --pgxs)
 include $(PGXS)