Klrp timings
                From: Table sent by A. Ramsak, May 31, 1999


Summary

    (1) The code speed is correlated with the THEORETICAL MFLOPS
        speed, not with clock speed. (Sun results are not
	definitive because FPU may not have run at full speed.)
        Compiler options are important.


Table I.  klrp (A. Ramsak).  RC is 10**5 divided by the CPU time and clock 
frequency. RF is 10**5 divided by the CPU time and theoretical MFLOPS 
rating (which is clock * FP pipeline).
--------------------------------------------------------------------------
                               Time           RC       RF
--------------------------------------------------------------------------
SGI PCL R8000 75-90 MHz *1)
    f77 -Ofast=ip21 -O3 
        -pfa -WK,-p=1           235          5.1      1.3
    f77 -Ofast=ip21 -O3 
	 -pfa -WK,-p=1
         -TENV:X=4 -LNO:ou=4    235          5.1      1.3
    f77 -Ofast=ip21  -O3        251          4.8      1.2
    f77 -r8000 -mips4 -n32 -O3 
         -pfa -WK,-p=1          479          2.5      0.6
    f77 -r8000 -mips4 -n32 -O3
                                599          2.0      0.5

Sun Enterprise 4500 336 MHz *2)
    f77 -fast
    f77 -fast -O5
    f77 -fast -O5 
        -xtarget=ultra2 
	-xcache=16/32/1:
	    4096/64/1           211          1.4      0.7 *2)

Intel Pentium II 400 MHz
    pgf77 $FLAGS                151          1.7      1.7
--------------------------------------------------------------------------
*1)   RC, RF calculated with an effective clock of 83 MHz because of 
    lack of information on which processor the job was run.
*2)   FPU unit may not have run at optimum speed.