* Program: STREAM_TUNED * Programmer: John D. McCalpin * Revision: 1.0, November 6, 2002 * * STREAM measures memory transfer rates in MB/s for simple * computational kernels coded in Fortran. * The intent is to demonstrate the extent to which ordinary user * code can exploit the main memory bandwidth of the system under * test. * * This version is a simple harness to allow code optimization * in the context of the data flow and result checking of the * basic STREAM version 5.0 code. Each of the four kernel loops * has been moved to a separate subroutine to allow easy code * optimization or replacement. * *========================================================================= * THIS IS JUST A STARTING POINT --- IT HAS NOT BEEN OPTIMIZED YET!!! *========================================================================= * The STREAM web page is at: * http://www.streambench.org * * Most of the content is currently hosted at: * http://www.cs.virginia.edu/stream/ * * BRIEF INSTRUCTIONS: * 0) See http://www.cs.virginia.edu/stream/ref.html for details * 1) STREAM requires a timing function called second(). * Several examples are provided in this directory. * "CPU" timers are only allowed for uniprocessor runs. * "Wall-clock" timers are required for all multiprocessor runs. * 2) The STREAM array sizes must be set to size the test. * The value "N" must be chosen so that each of the three * arrays is at least 4x larger than the sum of all the last- * level caches used in the run, or 1 million elements, which- * ever is larger. * ------------------------------------------------------------ * Note that you are free to use any array length and offset * that makes each array 4x larger than the last-level cache. * The intent is to determine the *best* sustainable bandwidth * available with this simple coding. Of course, lower values * are usually fairly easy to obtain on cached machines, but * by keeping the test to the *best* results, the answers are * easier to interpret. * You may put the arrays in common or not, at your discretion. * There is a commented-out COMMON statement below. * Fortran90 "allocatable" arrays are fine, too. * ------------------------------------------------------------ * 3) Compile the code with full optimization. Many compilers * generate unreasonably bad code before the optimizer tightens * things up. If the results are unreasonably good, on the * other hand, the optimizer might be too smart for me * Please let me know if this happens. * 4) Mail the results to mccalpin@cs.virginia.edu * Be sure to include: * a) computer hardware model number and software revision * b) the compiler flags * c) all of the output from the test case. * Please let me know if you do not want your name posted along * with the submitted results. * 5) See the web page for more comments about the run rules and * about interpretation of the results. * * Thanks, * Dr. Bandwidth *========================================================================= * PROGRAM stream * IMPLICIT NONE C .. Parameters .. INTEGER n,offset,ndim,ntimes PARAMETER (n=2000000,offset=0,ndim=n+offset,ntimes=10) C .. C .. Local Scalars .. DOUBLE PRECISION dummy,scalar,t INTEGER j,k,nbpw,quantum C .. C .. Local Arrays .. DOUBLE PRECISION maxtime(4),mintime(4),avgtime(4), $ times(4,ntimes) INTEGER bytes(4) CHARACTER label(4)*11 C .. C .. External Functions .. DOUBLE PRECISION second INTEGER checktick,realsize EXTERNAL second,checktick,realsize C .. C .. Intrinsic Functions .. C INTRINSIC dble,max,min,nint,sqrt C .. C .. Arrays in Common .. DOUBLE PRECISION a(ndim),b(ndim),c(ndim) C .. C .. Common blocks .. * COMMON a,b,c C .. C .. Data statements .. DATA avgtime/4*0.0D0/,mintime/4*1.0D+36/,maxtime/4*0.0D0/ DATA label/'Copy: ','Scale: ','Add: ', $ 'Triad: '/ DATA bytes/2,2,3,3/,dummy/0.0d0/ C .. * --- SETUP --- determine precision and check timing --- nbpw = realsize() WRITE (*,FMT=9010) 'Array size = ',n WRITE (*,FMT=9010) 'Offset = ',offset WRITE (*,FMT=9020) 'The total memory requirement is ', $ 3*nbpw*n/ (1024*1024),' MB' WRITE (*,FMT=9030) 'You are running each test ',ntimes,' times' WRITE (*,FMT=9030) '--' WRITE (*,FMT=9030) 'The *best* time for each test is used' WRITE (*,FMT=9030) '*EXCLUDING* the first and last iterations' !$OMP PARALLEL DO DO 10 j = 1,n a(j) = 2.0d0 b(j) = 0.5D0 c(j) = 0.0D0 10 CONTINUE t = second(dummy) !$OMP PARALLEL DO DO 20 j = 1,n a(j) = 0.5d0*a(j) 20 CONTINUE t = second(dummy) - t PRINT *,'----------------------------------------------------' quantum = checktick() WRITE (*,FMT=9000) $ 'Your clock granularity/precision appears to be ',quantum, $ ' microseconds' PRINT *,'----------------------------------------------------' * --- MAIN LOOP --- repeat test cases NTIMES times --- scalar = 0.5d0*a(1) DO 70 k = 1,ntimes t = second(dummy) a(1) = a(1) + t call stream_copy (c, a, n) t = second(dummy) - t c(n) = c(n) + t times(1,k) = t t = second(dummy) c(1) = c(1) + t call stream_scale (b, c, scalar, n) t = second(dummy) - t b(n) = b(n) + t times(2,k) = t t = second(dummy) a(1) = a(1) + t call stream_add (c, a, b, n) t = second(dummy) - t c(n) = c(n) + t times(3,k) = t t = second(dummy) b(1) = b(1) + t call stream_triad (a, b, c, scalar, n) t = second(dummy) - t a(n) = a(n) + t times(4,k) = t 70 CONTINUE * --- SUMMARY --- DO 90 k = 2,ntimes-1 DO 80 j = 1,4 avgtime(j) = avgtime(j) + times(j,k) mintime(j) = min(mintime(j),times(j,k)) maxtime(j) = max(maxtime(j),times(j,k)) 80 CONTINUE 90 CONTINUE WRITE (*,FMT=9040) DO 100 j = 1,4 avgtime(j) = avgtime(j)/dble(ntimes-2) WRITE (*,FMT=9050) label(j),n*bytes(j)*nbpw/mintime(j)/1.0D6, $ avgtime(j),mintime(j),maxtime(j) 100 CONTINUE PRINT *,'----------------------------------------------------' CALL checksums (a,b,c,n,ntimes) PRINT *,'----------------------------------------------------' 9000 FORMAT (1x,a,i6,a) 9010 FORMAT (1x,a,i10) 9020 FORMAT (1x,a,i4,a) 9030 FORMAT (1x,a,i3,a,a) 9040 FORMAT ('Function',5x,'Rate (MB/s) Avg time Min time Max time' $ ) 9050 FORMAT (a,4 (f10.4,2x)) END *------------------------------------- * INTEGER FUNCTION dblesize() * * A semi-portable way to determine the precision of DOUBLE PRECISION * in Fortran. * Here used to guess how many bytes of storage a DOUBLE PRECISION * number occupies. * INTEGER FUNCTION realsize() * IMPLICIT NONE C .. Local Scalars .. DOUBLE PRECISION result,test INTEGER j,ndigits C .. C .. Local Arrays .. DOUBLE PRECISION ref(30) C .. C .. External Subroutines .. EXTERNAL confuse C .. C .. Intrinsic Functions .. INTRINSIC abs,acos,log10,sqrt C .. C Test #1 - compare single(1.0d0+delta) to 1.0d0 10 DO 20 j = 1,30 ref(j) = 1.0d0 + 10.0d0** (-j) 20 CONTINUE DO 30 j = 1,30 test = ref(j) ndigits = j CALL confuse(test,result) IF (test.EQ.1.0D0) THEN GO TO 40 END IF 30 CONTINUE GO TO 50 40 WRITE (*,FMT='(a)') $ '----------------------------------------------' WRITE (*,FMT='(1x,a,i2,a)') 'Double precision appears to have ', $ ndigits,' digits of accuracy' IF (ndigits.LE.8) THEN realsize = 4 ELSE realsize = 8 END IF WRITE (*,FMT='(1x,a,i1,a)') 'Assuming ',realsize, $ ' bytes per DOUBLE PRECISION word' WRITE (*,FMT='(a)') $ '----------------------------------------------' RETURN 50 PRINT *,'Hmmmm. I am unable to determine the size.' PRINT *,'Please enter the number of Bytes per DOUBLE PRECISION', $ ' number : ' READ (*,FMT=*) realsize IF (realsize.NE.4 .AND. realsize.NE.8) THEN PRINT *,'Your answer ',realsize,' does not make sense.' PRINT *,'Try again.' PRINT *,'Please enter the number of Bytes per ', $ 'DOUBLE PRECISION number : ' READ (*,FMT=*) realsize END IF PRINT *,'You have manually entered a size of ',realsize, $ ' bytes per DOUBLE PRECISION number' WRITE (*,FMT='(a)') $ '----------------------------------------------' END SUBROUTINE confuse(q,r) * IMPLICIT NONE C .. Scalar Arguments .. DOUBLE PRECISION q,r C .. C .. Intrinsic Functions .. INTRINSIC cos C .. r = cos(q) RETURN END * A semi-portable way to determine the clock granularity * Adapted from a code by John Henning of Digital Equipment Corporation * INTEGER FUNCTION checktick() * IMPLICIT NONE C .. Parameters .. INTEGER n PARAMETER (n=20) C .. C .. Local Scalars .. DOUBLE PRECISION dummy,t1,t2 INTEGER i,j,jmin C .. C .. Local Arrays .. DOUBLE PRECISION timesfound(n) C .. C .. External Functions .. DOUBLE PRECISION second EXTERNAL second C .. C .. Intrinsic Functions .. INTRINSIC max,min,nint C .. i = 0 dummy = 0.0d0 t1 = second(dummy) 10 t2 = second(dummy) IF (t2.EQ.t1) GO TO 10 t1 = t2 i = i + 1 timesfound(i) = t1 IF (i.LT.n) GO TO 10 jmin = 1000000 DO 20 i = 2,n j = nint((timesfound(i)-timesfound(i-1))*1d6) jmin = min(jmin,max(j,0)) 20 CONTINUE IF (jmin.GT.0) THEN checktick = jmin ELSE PRINT *,'Your clock granularity appears to be less ', $ 'than one microsecond' checktick = 1 END IF RETURN * PRINT 14, timesfound(1)*1d6 * DO 20 i=2,n * PRINT 14, timesfound(i)*1d6, * & nint((timesfound(i)-timesfound(i-1))*1d6) * 14 FORMAT (1X, F18.4, 1X, i8) * 20 CONTINUE END SUBROUTINE checksums(a,b,c,n,ntimes) * IMPLICIT NONE C .. C .. Arguments .. DOUBLE PRECISION a(*),b(*),c(*) INTEGER n,ntimes C .. C .. Local Scalars .. DOUBLE PRECISION aa,bb,cc,scalar,suma,sumb,sumc,epsilon INTEGER k C .. C Repeat the main loop, but with scalars only. C This is done to check the sum & make sure all C iterations have been executed correctly. aa = 2.0D0 bb = 0.5D0 cc = 0.0D0 aa = 0.5D0*aa scalar = 0.5d0*aa DO k = 1,ntimes cc = aa bb = scalar*cc cc = aa + bb aa = bb + scalar*cc END DO aa = aa*DBLE(n-2) bb = bb*DBLE(n-2) cc = cc*DBLE(n-2) C Now sum up the arrays, excluding the first and last C elements, which are modified using the timing results C to confuse aggressive optimizers. suma = 0.0d0 sumb = 0.0d0 sumc = 0.0d0 !$OMP PARALLEL DO REDUCTION(+:suma,sumb,sumc) DO 110 j = 2,n-1 suma = suma + a(j) sumb = sumb + b(j) sumc = sumc + c(j) 110 CONTINUE epsilon = 1.D-6 IF (ABS(suma-aa)/suma .GT. epsilon) THEN PRINT *,'Failed Validation on array a()' PRINT *,'Target Sum of a is = ',aa PRINT *,'Computed Sum of a is = ',suma ELSEIF (ABS(sumb-bb)/sumb .GT. epsilon) THEN PRINT *,'Failed Validation on array b()' PRINT *,'Target Sum of b is = ',bb PRINT *,'Computed Sum of b is = ',sumb ELSEIF (ABS(sumc-cc)/sumc .GT. epsilon) THEN PRINT *,'Failed Validation on array c()' PRINT *,'Target Sum of c is = ',cc PRINT *,'Computed Sum of c is = ',sumc ELSE PRINT *,'Solution Validates!' ENDIF END subroutine stream_copy (c, a, n) real*8 c(*), a(*) !$OMP PARALLEL DO do j = 1,n c(j) = a(j) end do end subroutine stream_scale (b, c, scalar, n) real*8 b(*), c(*), scalar !$OMP PARALLEL DO do j = 1,n b(j) = scalar*c(j) end do end subroutine stream_add (c, a, b, n) real*8 c(*), a(*), b(*) !$OMP PARALLEL DO do j = 1,n c(j) = a(j) + b(j) end do end subroutine stream_triad (a, b, c, scalar, n) real*8 a(*), b(*), c(*), scalar !$OMP PARALLEL DO do j = 1,n a(j) = b(j) + scalar*c(j) end do end