########################################################################
#
# environment variable CPU is used to set platform-dependent make variables.
#
# If CPU is undefined; use the test_os script to figure it out:
ifeq ($(CPU),)
   CPU := $(shell ./test_os)
endif

# Od Dec alpha comiler options:

#  FC = f77
# FFLAGS = -O -fast
#  FFLAGS = -O -fast -r8

# Sun compiler options which Bowman found is fastest, particularly for
# inlining square roots:

# FFLAGS = -fast -O4 -Bstatic

# DEC alpha Linux options:
ifeq ($(CPU),LINUX_alpha)
  FC = fort
  FFLAGS =  -extend_source -r8 -fast -assume no2underscores
endif

ifeq ($(CPU),LINUX_lahey)
  FC = lf95
  FFLAGS   = --dbl -O --tpp --wide --nap --nchk --npca --nsav --ntrace \
    --prefetch 2 --staticlink --info
# --mldefault cdecl
#   --prefetch 2 --staticlink -X9
#   --prefetch 2 --staticlink -L/usr/local/lff95/lib -lblas
#   FFLAGS = -O2
endif

# Don't want inlining with interprocedural optimization turned on for 
# these benchmarking codes, because the compiler may figure out that the 
# external loops only need to be done once...

ifeq ($(CPU),LINUX_INTEL)
  FC = ifc
  FFLAGS   = -r8 -O3 -tpp7 -axiMKW -pad -unroll
# In some case -pc32 can speed up / or sqrt ops
#  -Zp16 or -Zp8 had no effect
# FFLAGS   = -r8 -O3 -tpp7 -axiMKW -pad -unroll -fno-alias -fno-fnalias
# alias switches had no effect
# FFLAGS   = -r8 -O3 -tpp7 -axiMKW -pad -unroll -parallel
endif

all: stream stream_tuned

stream:	stream_d.f mysecond.f
	$(FC) $(FFLAGS) -o stream stream_d.f mysecond.f

stream_tuned:	stream_tuned.f second.f
	$(FC) $(FFLAGS) -o stream_tuned stream_tuned.f second.f

speed:	speed.f speedsub.f
	$(FC) $(FFLAGS) -c speedsub.f saxpy_g.f
#	$(FC) $(FFLAGS) -o speed speed.f speedsub.o saxpy_g.o
	$(FC) $(FFLAGS) -o speed speed.f speedsub.o saxpy_g.o \
              /usr/local/lff95/lib/libblas.a
#             $(F90HOME)/lib/libblas.a  # Doesn't exist yet
#
# Have to force link to blas??
#
#	$(FC) $(FFLAGS) -c speedsub.f second.f
#	$(FC) $(FFLAGS) -o speed speed.f speedsub.o second.o blas/libblas1.a
# The free pentium-optimized assembly-language blas from 1996 in 
# subdirectory blas are  
# 15% faster for small vectors but
# 30% slower for large vectors.  So don't bother with them.

linpacks: linpacks.f second.f
	$(FC) -O -o linpacks linpacks.f second.f

linpackd: linpackd.f second.f
	$(FC) -O -o linpackd linpackd.f second.f

whetstoned: whetstoned.f
	$(FC) -O -o whetstoned whetstoned.f

clock.o: clock.c
	cc -c clock.c

clean:
	rm -f *.o *~ stream speed stream_tuned