# run using
# 	poe <progname> -procs 4
# can also  use options
# 	-css_interrupt {yes | no}  (default no)
#	-eager_limit size_limit (is default $MP_EAGER_LIMIT above
#                                above this limit use rendezvous protocol)
#	-buffer_mem memsize (is default 2.8megs for IP 64meg US)
#	-single_thread {no | yes}  ? (default is no which means a 
#		it assumes multiple message passing threads and
#		adds lock overhead.  If no thread, then say 'yes'
# can use cpus_per_node and nodes together instead of -procs
#	-cpu_use numcpuspernode (default is $MP_TASKS_PER_NODE)
#	-nodes numnodesused 
#	-euidevice {en0 | fi0 | tr0 | css0} (default in $MP_EUIDEVICE)
#	-euilib {ip | us}  (default in $MP_EUILIB)
CC = mpcc
CXX = mpCC
CFLAGS = -O3 -qarch=pwr3 -qansialias -qlist -qsource -DSP2 -DNITER=3000
MPIHOME = /usr/local
MPIINC = -I${MPIHOME}
MPILIB = -L${MPIHOME} -lm
FLEXIO = /u4/jshalf
FLEXINC = -I$(FLEXIO)/include
FLEXLIB = -L$(FLEXIO)/lib -lieeeio
# SerialTest  : Test the WaveToy stuff 
# ParTest  : Just test the domain decomp machinery
# MPItest : test simplest MPI comm
# PthreadTest : test cube domain decomposition with pthreads
# PthreadSlice : test slice domain decomp with pthreads
MAIN = MPIperfExchange
# MAIN = PthreadTest

LD = -lpthread

include Makefile.rules


# Tests: (prelim) MPIperf (regular MPI isend)
#####################################################
# 10k iter
# 8 procs: 46:8203	2 procs: 36.073
# 3000 iter
# 8 procs: 13.752	2 procs: 9.91353
# cssinterrupt
# 8 procs: 15.5324
# 2 procs: 10.0334
# single_thread yes 
# 8 procs: 14.1984
# 2 procs: 9.86464
# eagerlimit 0
# 8 procs: 17.8428
# eagerlimit 20000 (couldn't push it futher)
# 8 procs: 15.089 seconds
# euilib ip
# 8 procs: 26.5896
# 2 procs: 17.8684
# buffer mem 3000000
# 8 procs: 13.842
# buffer mem 30000000
# 8 procs: 13.4472