SGI F77 COMPILER BASIC TEST - O2 R1000 175 MHz
			      Updated 99/04/13


OBJECTIVE

First tests on O2/R10000: to comaper with Origin 2000.


DESCRIPTION

Machine: O2, R10000, IP32: 175 MHz (350 MFLOPS).
Data cache size: 32 Kbytes
Instruction cache size: 32 Kbytes
Secondary unified instruction/data cache size: 1 Mbyte on Processor 0
Main memory size: 128 Mbytes

The machine was running a normal multi-desktop X-windows session 
during the tests.

Program: Matrix multiplication using nested DO loops.
Matrix dimensions: ND = 800 (reservation).
Timing: dtime (tempd.f)
Program details:
    P25:    I-loop inside (vector style), no directives:

      PROGRAM P25
      ...
      IMPLICIT    REAL*8  (A-H,O-Z)
      PARAMETER   (ND = 800, NIN = 5, NOUT = 6)
      DIMENSION   A(ND,ND), B(ND,ND), C(ND,ND)
      ...
      DO  26  J = 1,N
          DO  24  K = 1,N
              DO  22  I = 1,N
                  C(I,J) = C(I,J) + A(I,K) * B(K,J)
   22         CONTINUE
   24     CONTINUE
   26 CONTINUE
      ...


    P25V:    calls DGEMM


RESULTS

Clearly this is not primarily a computational machine, its best results
are about 67% compared to over 80% on other SGI R8000 and R10000 machines.


Table I.  ND = 800, N = 800. Theoretical CPU time lower limit is 2.9
seconds (175 MHz R10000).
--------------------------------------------------------------------------
Program   Compiler call                               Threads  CPU time
--------------------------------------------------------------------------
p25       f77 -Ofast=ip32                                        6.7
              -LNO:ou=2                                          8.7
              -LNO:ou=4                                          7.0

          f77 -r10000 -mips4 -n32 -O3                            6.4  *1
              -LNO:ou=2                                          8.7
              -LNO:ou=4                                          6.9
              -LNO:ou=6                                          7.5

p25v      f77 -Ofast=ip32 -lcomplib.sgimath                      4.4  *2
          f77 -r10000 -mips4 -n32 -O3 -lcomplib.sgimath          4.4
--------------------------------------------------------------------------
*1  R = 2.9/6.4 = 0.46, while on R8000 the best R without DGEMM was 0.53 
    using advanced options.
*2  R = 2.9/4.4 = 0.66.



Table II.  ND = 1600, N = 1600. Theoretical CPU time lower limit is 23.4
seconds (175 MHz R10000). There was no swapping activity associated with 
the main calculation.
--------------------------------------------------------------------------
Program   Compiler call                               Threads  CPU time
--------------------------------------------------------------------------
p25       f77 -Ofast=ip32                                       54    *1
              -LNO:ou=2                                        197
              -LNO:ou=4                                         93
              -LNO:ou=6                                         79

          f77 -r10000 -mips4 -n32 -O3                           53.7  *1
              -LNO:ou=2                                        188
              -LNO:ou=4                                         93
              -LNO:ou=6                                         78

p25v      f77 -Ofast=ip32 -lcomplib.sgimath                     35.0  *2
          f77 -r10000 -mips4 -n32 -O3 -lcomplib.sgimath         35.4
--------------------------------------------------------------------------
*1  R = 23.4/54 = 0.43, while on R8000 the best R without DGEMM was 0.53 
    using advanced options.
*2  R = 23.4/35.0 = 0.67.