Merge ndk/machinefiles/depends-intel-fpmodelfast into master (PR #1761)

For Intel builds, there are 7 HOMME fortran files we build with higher optimization flags.
Add -fp-model fast to those 7 builds as this avoids the slow pow function on KNL, and may
generally improve peformance for those files.
Note that the base flags for intel add -fp-model source and I have verified that the compiler will favor "fast" when compiled with BOTH -fp-model source and -fp-model fast.
Passes acme_developer
Also added a set of depends for Intel v 18.

Fixes #1698

[non-BFB]   roundoff level changes

config/acme/machines/Depends.intel

@@ -30,7 +30,7 @@ kissvec.o
 
 ifeq ($(DEBUG),FALSE)
   $(PERFOBJS): %.o: %.F90
-	  $(FC) -c $(INCLDIR) $(INCS) $(FFLAGS) $(FREEFLAGS)  -O3  -no-prec-div $<
+	  $(FC) -c $(INCLDIR) $(INCS) $(FFLAGS) $(FREEFLAGS)  -O3 -fp-model fast -no-prec-div $<
   $(REDUCED_OPT_OBJS): %.o: %.F90
 	  $(FC) -c $(INCLDIR) $(INCS) $(FFLAGS) $(FREEFLAGS)  -O1 $<
   $(REDUCED_PRECISION_OBJS): %.o: %.F90

config/acme/machines/Depends.intel18

@@ -0,0 +1,36 @@
+#
+PERFOBJS=\
+prim_advection_mod_base.o \
+vertremap_mod_base.o \
+edge_mod_base.o \
+derivative_mod_base.o \
+bndry_mod_base.o \
+prim_advance_mod.o \
+uwshcu.o
+
+# shr_wv_sat_mod does not need to have better than ~0.1% precision, and benefits
+# enormously from a lower precision in the vector functions.
+REDUCED_PRECISION_OBJS=\
+shr_wv_sat_mod.o
+
+SHR_RANDNUM_FORT_OBJS=\
+kissvec_mod.o \
+mersennetwister_mod.o \
+dSFMT_interface.o \
+shr_RandNum_mod.o
+
+SHR_RANDNUM_C_OBJS=\
+dSFMT.o \
+dSFMT_utils.o \
+kissvec.o
+
+ifeq ($(DEBUG),FALSE)
+  $(PERFOBJS): %.o: %.F90
+	  $(FC) -c $(INCLDIR) $(INCS) $(FFLAGS) $(FREEFLAGS)  -O3 -fp-model fast -no-prec-div $<
+  $(REDUCED_PRECISION_OBJS): %.o: %.F90
+	  $(FC) -c $(INCLDIR) $(INCS) $(FFLAGS) $(FREEFLAGS) -fimf-precision=low -fp-model fast $<
+  $(SHR_RANDNUM_FORT_OBJS): %.o: %.F90
+	  $(FC) -c $(INCLDIR) $(INCS) $(FFLAGS) $(FREEFLAGS) -O3 -fp-model fast -no-prec-div -no-prec-sqrt -qoverride-limits $<
+  $(SHR_RANDNUM_C_OBJS): %.o: %.c
+	  $(CC) -c $(INCLDIR) $(INCS) $(CFLAGS) -O3 -fp-model fast $<
+endif