1 2 # include <stdio.h> 3 # include <math.h> 4 # include <limits.h> 5 # include <float.h> 6 7 /* 8 * Program: Stream 9 * Programmer: Joe R. Zagar 10 * Revision: 4.0-BETA, October 24, 1995 11 * Original code developed by John D. McCalpin 12 * 13 * This program measures memory transfer rates in MB/s for simple 14 * computational kernels coded in C. These numbers reveal the quality 15 * of code generation for simple uncacheable kernels as well as showing 16 * the cost of floating-point operations relative to memory accesses. 17 * 18 * INSTRUCTIONS: 19 * 20 * 1) Stream requires a good bit of memory to run. Adjust the 21 * value of 'N' (below) to give a 'timing calibration' of 22 * at least 20 clock-ticks. This will provide rate estimates 23 * that should be good to about 5% precision. 24 */ 25 26 # define N 2000000 27 # define NTIMES 50 28 # define OFFSET 0 29 30 /* 31 * 3) Compile the code with full optimization. Many compilers 32 * generate unreasonably bad code before the optimizer tightens 33 * things up. If the results are unreasonably good, on the 34 * other hand, the optimizer might be too smart for me! 35 * 36 * Try compiling with: 37 * cc -O stream_d.c second.c -o stream_d -lm 38 * 39 * This is known to work on Cray, SGI, IBM, and Sun machines. 40 * 41 * 42 * 4) Mail the results to mccalpin@cs.virginia.edu 43 * Be sure to include: 44 * a) computer hardware model number and software revision 45 * b) the compiler flags 46 * c) all of the output from the test case. 47 * Thanks! 48 * 49 */ 50 51 # define HLINE "-------------------------------------------------------------\n" 52 53 # ifndef MIN 54 # define MIN(x,y) ((x)<(y) ? (x) : (y)) 55 # endif 56 # ifndef MAX 57 # define MAX(x,y) ((x)>(y) ? (x) : (y)) 58 # endif 59 60 static double a[N+OFFSET], 61 b[N+OFFSET], 62 c[N+OFFSET]; 63 /*double *a,*b,*c;*/ 64 65 static double mintime[4] = {FLT_MAX,FLT_MAX,FLT_MAX,FLT_MAX}; 66 67 static const char *label[4] = {"Copy: ", "Scale: ", "Add: ", "Triad: "}; 68 69 static double bytes[4] = { 70 2 * sizeof(double) * N, 71 2 * sizeof(double) * N, 72 3 * sizeof(double) * N, 73 3 * sizeof(double) * N 74 }; 75 76 #include <petscsys.h> 77 78 int main(int argc,char **args) 79 { 80 int quantum, checktick(void); 81 register int j, k; 82 double scalar, t, times[4][NTIMES],irate[4],rate[4]; 83 int rank,size,resultlen; 84 char hostname[MPI_MAX_PROCESSOR_NAME]; 85 MPI_Status status; 86 87 MPI_Init(&argc,&args); 88 MPI_Comm_rank(MPI_COMM_WORLD,&rank); 89 MPI_Comm_size(MPI_COMM_WORLD,&size); 90 if (!rank) printf("Number of MPI processes %d\n",size); 91 92 for (j=0; j<MPI_MAX_PROCESSOR_NAME; j++) { 93 hostname[j] = 0; 94 } 95 MPI_Get_processor_name(hostname,&resultlen); 96 if (!rank) { 97 printf("Process %d %s\n",rank,hostname); 98 for (j=1; j<size; j++) { 99 MPI_Recv(hostname,MPI_MAX_PROCESSOR_NAME,MPI_CHAR,j,0,MPI_COMM_WORLD,&status); 100 printf("Process %d %s\n",j,hostname); 101 } 102 } else { 103 MPI_Send(hostname,MPI_MAX_PROCESSOR_NAME,MPI_CHAR,0,0,MPI_COMM_WORLD); 104 } 105 MPI_Barrier(MPI_COMM_WORLD); 106 107 /* --- SETUP --- determine precision and check timing --- */ 108 109 if (!rank) { 110 /*printf(HLINE); 111 printf("Array size = %d, Offset = %d\n" , N, OFFSET); 112 printf("Total memory required = %.1f MB.\n", (3 * N * BytesPerWord) / 1048576.0); 113 printf("Each test is run %d times, but only\n", NTIMES); 114 printf("the *best* time for each is used.\n"); 115 printf(HLINE); */ 116 } 117 118 /* Get initial value for system clock. */ 119 120 /* a = malloc(N*sizeof(double)); 121 b = malloc(N*sizeof(double)); 122 c = malloc(N*sizeof(double));*/ 123 for (j=0; j<N; j++) { 124 a[j] = 1.0; 125 b[j] = 2.0; 126 c[j] = 0.0; 127 } 128 129 if (!rank) { 130 if ((quantum = checktick()) >= 1) ; /* printf("Your clock granularity/precision appears to be %d microseconds.\n", quantum); */ 131 else ; /* printf("Your clock granularity appears to be less than one microsecond.\n");*/ 132 } 133 134 t = MPI_Wtime(); 135 for (j = 0; j < N; j++) a[j] = 2.0E0 * a[j]; 136 t = 1.0E6 * (MPI_Wtime() - t); 137 138 if (!rank) { 139 /* printf("Each test below will take on the order of %d microseconds.\n", (int) t); 140 printf(" (= %d clock ticks)\n", (int) (t/quantum)); 141 printf("Increase the size of the arrays if this shows that\n"); 142 printf("you are not getting at least 20 clock ticks per test.\n"); 143 printf(HLINE);*/ 144 } 145 146 147 /* --- MAIN LOOP --- repeat test cases NTIMES times --- */ 148 149 scalar = 3.0; 150 for (k=0; k<NTIMES; k++) 151 { 152 MPI_Barrier(MPI_COMM_WORLD); 153 times[0][k] = MPI_Wtime(); 154 /* should all these barriers be pulled outside of the time call? */ 155 MPI_Barrier(MPI_COMM_WORLD); 156 for (j=0; j<N; j++) c[j] = a[j]; 157 MPI_Barrier(MPI_COMM_WORLD); 158 times[0][k] = MPI_Wtime() - times[0][k]; 159 160 times[1][k] = MPI_Wtime(); 161 MPI_Barrier(MPI_COMM_WORLD); 162 for (j=0; j<N; j++) b[j] = scalar*c[j]; 163 MPI_Barrier(MPI_COMM_WORLD); 164 times[1][k] = MPI_Wtime() - times[1][k]; 165 166 times[2][k] = MPI_Wtime(); 167 MPI_Barrier(MPI_COMM_WORLD); 168 for (j=0; j<N; j++) c[j] = a[j]+b[j]; 169 MPI_Barrier(MPI_COMM_WORLD); 170 times[2][k] = MPI_Wtime() - times[2][k]; 171 172 times[3][k] = MPI_Wtime(); 173 MPI_Barrier(MPI_COMM_WORLD); 174 for (j=0; j<N; j++) a[j] = b[j]+scalar*c[j]; 175 MPI_Barrier(MPI_COMM_WORLD); 176 times[3][k] = MPI_Wtime() - times[3][k]; 177 } 178 179 /* --- SUMMARY --- */ 180 181 for (k=0; k<NTIMES; k++) 182 for (j=0; j<4; j++) mintime[j] = MIN(mintime[j], times[j][k]); 183 184 for (j=0; j<4; j++) irate[j] = 1.0E-06 * bytes[j]/mintime[j]; 185 MPI_Reduce(irate,rate,4,MPI_DOUBLE,MPI_SUM,0,MPI_COMM_WORLD); 186 187 if (!rank) { 188 printf("Function Rate (MB/s) \n"); 189 for (j=0; j<4; j++) printf("%s%11.4f\n", label[j],rate[j]); 190 } 191 MPI_Finalize(); 192 return 0; 193 } 194 195 # define M 20 196 197 int checktick(void) 198 { 199 int i, minDelta, Delta; 200 double t1, t2, timesfound[M]; 201 202 /* Collect a sequence of M unique time values from the system. */ 203 204 for (i = 0; i < M; i++) { 205 t1 = MPI_Wtime(); 206 while (((t2=MPI_Wtime()) - t1) < 1.0E-6) ; 207 timesfound[i] = t1 = t2; 208 } 209 210 /* 211 * Determine the minimum difference between these M values. 212 * This result will be our estimate (in microseconds) for the 213 * clock granularity. 214 */ 215 216 minDelta = 1000000; 217 for (i = 1; i < M; i++) { 218 Delta = (int)(1.0E6 * (timesfound[i]-timesfound[i-1])); 219 minDelta = MIN(minDelta, MAX(Delta,0)); 220 } 221 222 return(minDelta); 223 } 224 225