xref: /petsc/src/benchmarks/streams/SSEVersion.c (revision 7b6bb2c608b6fc6714ef38fda02c2dbb91c82665)
1 static const char help[] = "STREAM benchmark specialized for SSE2\n\\n";
2 
3 /* Note: this file has been modified significantly from its original version */
4 #include <emmintrin.h>
5 #include <petsctime.h>
6 #include <petscsys.h>
7 #include <numa.h>
8 #include <limits.h>
9 #include <float.h>
10 
11 #ifndef SSE2
12 #  define SSE2 1
13 #endif
14 #ifndef __SSE2__
15 #  error SSE2 instruction set is not enabled, try adding -march=native to CFLAGS or disable by adding -DSSE2=0
16 #endif
17 #ifndef PREFETCH_NTA /* Use software prefetch and set non-temporal policy so that lines evicted from L1D will not subsequently reside in L2 or L3. */
18 #  define PREFETCH_NTA 1
19 #endif
20 #ifndef STATIC_ALLOC /* Statically allocate the vectors. Most platforms do not find physical pages when memory is allocated, therefore the faulting strategy still affects performance. */
21 #  define STATIC_ALLOC 0
22 #endif
23 #ifndef FAULT_TOGETHER /* Faults all three vectors together which usually interleaves DRAM pages in physical memory. */
24 #  define FAULT_TOGETHER 0
25 #endif
26 #ifndef USE_MEMCPY /* Literally call memcpy(3) for the COPY benchmark. Some compilers detect the unoptimized loop as memcpy and call this anyway. */
27 #  define USE_MEMCPY 0
28 #endif
29 
30 /*
31  * Program: Stream
32  * Programmer: Joe R. Zagar
33  * Revision: 4.0-BETA, October 24, 1995
34  * Original code developed by John D. McCalpin
35  *
36  * This program measures memory transfer rates in MB/s for simple
37  * computational kernels coded in C.  These numbers reveal the quality
38  * of code generation for simple uncacheable kernels as well as showing
39  * the cost of floating-point operations relative to memory accesses.
40  *
41  * INSTRUCTIONS:
42  *
43  *	1) Stream requires a good bit of memory to run.  Adjust the
44  *          value of 'N' (below) to give a 'timing calibration' of
45  *          at least 20 clock-ticks.  This will provide rate estimates
46  *          that should be good to about 5% precision.
47  */
48 
49 # define N      4000000
50 # define NTIMES	100
51 # define OFFSET	0
52 
53 # define HLINE "-------------------------------------------------------------\n"
54 
55 # ifndef MIN
56 # define MIN(x,y) ((x)<(y)?(x):(y))
57 # endif
58 # ifndef MAX
59 # define MAX(x,y) ((x)>(y)?(x):(y))
60 # endif
61 
62 #if STATIC_ALLOC
63   double a[N+OFFSET],b[N+OFFSET],c[N+OFFSET];
64 #endif
65 
66 static int checktick(void);
67 static double Second(void);
68 
69 int main(int argc,char *argv[])
70 {
71   const char *label[4] = {"Copy", "Scale","Add", "Triad"};
72   const double bytes[4] = {2 * sizeof(double) * N,
73                            2 * sizeof(double) * N,
74                            3 * sizeof(double) * N,
75                            3 * sizeof(double) * N};
76   double rmstime[4] = {0},maxtime[4] = {0},mintime[4] = {FLT_MAX,FLT_MAX,FLT_MAX,FLT_MAX};
77   int	   quantum;
78   int	   BytesPerWord,j,k,size;
79   PetscInt node = -1;
80   double   scalar, t, times[4][NTIMES];
81 #if !STATIC_ALLOC
82   double   *PETSC_RESTRICT a,*PETSC_RESTRICT b,*PETSC_RESTRICT c;
83 #endif
84 
85   PetscInitialize(&argc,&argv,0,help);
86   MPI_Comm_size(PETSC_COMM_WORLD,&size);
87   PetscOptionsGetInt(PETSC_NULL,"-node",&node,PETSC_NULL);
88   /* --- SETUP --- determine precision and check timing --- */
89 
90   PetscPrintf(PETSC_COMM_WORLD,HLINE);
91   BytesPerWord = sizeof(double);
92   PetscPrintf(PETSC_COMM_WORLD,"This system uses %d bytes per DOUBLE PRECISION word.\n",
93          BytesPerWord);
94 
95   PetscPrintf(PETSC_COMM_WORLD,HLINE);
96   PetscPrintf(PETSC_COMM_WORLD,"Array size = %d, Offset = %d\n" , N, OFFSET);
97   PetscPrintf(PETSC_COMM_WORLD,"Total memory required = %.1f MB per process.\n",
98          (3 * N * BytesPerWord) / 1048576.0);
99   PetscPrintf(PETSC_COMM_WORLD,"Each test is run %d times, but only\n", NTIMES);
100   PetscPrintf(PETSC_COMM_WORLD,"the *best* time for each is used.\n");
101 
102   /* Get initial value for system clock. */
103 
104 #if !STATIC_ALLOC
105   if (node == -1) {
106     posix_memalign((void**)&a,64,N*sizeof(double));
107     posix_memalign((void**)&b,64,N*sizeof(double));
108     posix_memalign((void**)&c,64,N*sizeof(double));
109   } else if (node == -2) {
110     a = malloc(N*sizeof(double));
111     b = malloc(N*sizeof(double));
112     c = malloc(N*sizeof(double));
113   } else {
114     a = numa_alloc_onnode(N*sizeof(double),node);
115     b = numa_alloc_onnode(N*sizeof(double),node);
116     c = numa_alloc_onnode(N*sizeof(double),node);
117   }
118 #endif
119 #if FAULT_TOGETHER
120   for (j=0; j<N; j++) {
121     a[j] = 1.0;
122     b[j] = 2.0;
123     c[j] = 0.0;
124   }
125 #else
126   for (j=0; j<N; j++) a[j] = 1.0;
127   for (j=0; j<N; j++) b[j] = 2.0;
128   for (j=0; j<N; j++) c[j] = 0.0;
129 #endif
130 
131   PetscPrintf(PETSC_COMM_WORLD,HLINE);
132 
133   if  ( (quantum = checktick()) >= 1)
134     PetscPrintf(PETSC_COMM_WORLD,"Your clock granularity/precision appears to be "
135            "%d microseconds.\n", quantum);
136   else
137     PetscPrintf(PETSC_COMM_WORLD,"Your clock granularity appears to be "
138            "less than one microsecond.\n");
139 
140   t = Second();
141   for (j = 0; j < N; j++)
142     a[j] = 2.0E0 * a[j];
143   t = 1.0E6 * (Second() - t);
144 
145   PetscPrintf(PETSC_COMM_WORLD,"Each test below will take on the order"
146          " of %d microseconds.\n", (int) t  );
147   PetscPrintf(PETSC_COMM_WORLD,"   (= %d clock ticks)\n", (int) (t/quantum) );
148   PetscPrintf(PETSC_COMM_WORLD,"Increase the size of the arrays if this shows that\n");
149   PetscPrintf(PETSC_COMM_WORLD,"you are not getting at least 20 clock ticks per test.\n");
150 
151   PetscPrintf(PETSC_COMM_WORLD,HLINE);
152 
153   PetscPrintf(PETSC_COMM_WORLD,"WARNING -- The above is only a rough guideline.\n");
154   PetscPrintf(PETSC_COMM_WORLD,"For best results, please be sure you know the\n");
155   PetscPrintf(PETSC_COMM_WORLD,"precision of your system timer.\n");
156   PetscPrintf(PETSC_COMM_WORLD,HLINE);
157 
158   /*	--- MAIN LOOP --- repeat test cases NTIMES times --- */
159 
160   scalar = 3.0;
161   for (k=0; k<NTIMES; k++) {
162     MPI_Barrier(PETSC_COMM_WORLD);
163     /* ### COPY: c <- a ### */
164     times[0][k] = Second();
165     MPI_Barrier(PETSC_COMM_WORLD);
166 #if USE_MEMCPY
167     memcpy(c,a,N*sizeof(double));
168 #elif SSE2
169     for (j=0; j<N; j+=8) {
170       _mm_stream_pd(c+j+0,_mm_load_pd(a+j+0));
171       _mm_stream_pd(c+j+2,_mm_load_pd(a+j+2));
172       _mm_stream_pd(c+j+4,_mm_load_pd(a+j+4));
173       _mm_stream_pd(c+j+6,_mm_load_pd(a+j+6));
174 #  if PREFETCH_NTA
175       _mm_prefetch(a+j+64,_MM_HINT_NTA);
176 #  endif
177     }
178 #else
179     for (j=0; j<N; j++) c[j] = a[j];
180 #endif
181     MPI_Barrier(PETSC_COMM_WORLD);
182     times[0][k] = Second() - times[0][k];
183 
184     /* ### SCALE: b <- scalar * c ### */
185     times[1][k] = Second();
186     MPI_Barrier(PETSC_COMM_WORLD);
187 #if SSE2
188     {
189       __m128d scalar2 = _mm_set1_pd(scalar);
190       for (j=0; j<N; j+=8) {
191         _mm_stream_pd(b+j+0,_mm_mul_pd(scalar2,_mm_load_pd(c+j+0)));
192         _mm_stream_pd(b+j+2,_mm_mul_pd(scalar2,_mm_load_pd(c+j+2)));
193         _mm_stream_pd(b+j+4,_mm_mul_pd(scalar2,_mm_load_pd(c+j+4)));
194         _mm_stream_pd(b+j+6,_mm_mul_pd(scalar2,_mm_load_pd(c+j+6)));
195 #  if PREFETCH_NTA
196         _mm_prefetch(c+j+64,_MM_HINT_NTA);
197 #  endif
198       }
199     }
200 #else
201     for (j=0; j<N; j++) b[j] = scalar*c[j];
202 #endif
203     MPI_Barrier(PETSC_COMM_WORLD);
204     times[1][k] = Second() - times[1][k];
205 
206     /* ### ADD: c <- a + b ### */
207     times[2][k] = Second();
208     MPI_Barrier(PETSC_COMM_WORLD);
209 #if SSE2
210     {
211       for (j=0; j<N; j+=8) {
212         _mm_stream_pd(c+j+0,_mm_add_pd(_mm_load_pd(a+j+0),_mm_load_pd(b+j+0)));
213         _mm_stream_pd(c+j+2,_mm_add_pd(_mm_load_pd(a+j+2),_mm_load_pd(b+j+2)));
214         _mm_stream_pd(c+j+4,_mm_add_pd(_mm_load_pd(a+j+4),_mm_load_pd(b+j+4)));
215         _mm_stream_pd(c+j+6,_mm_add_pd(_mm_load_pd(a+j+6),_mm_load_pd(b+j+6)));
216 #  if PREFETCH_NTA
217         _mm_prefetch(a+j+64,_MM_HINT_NTA);
218         _mm_prefetch(b+j+64,_MM_HINT_NTA);
219 #  endif
220       }
221     }
222 #else
223     for (j=0; j<N; j++) c[j] = a[j]+b[j];
224 #endif
225     MPI_Barrier(PETSC_COMM_WORLD);
226     times[2][k] = Second() - times[2][k];
227 
228     /* ### TRIAD: a <- b + scalar * c ### */
229     times[3][k] = Second();
230     MPI_Barrier(PETSC_COMM_WORLD);
231 #if SSE2
232     {
233       __m128d scalar2 = _mm_set1_pd(scalar);
234       for (j=0; j<N; j+=8) {
235         _mm_stream_pd(a+j+0,_mm_add_pd(_mm_load_pd(b+j+0),_mm_mul_pd(scalar2,_mm_load_pd(c+j+0))));
236         _mm_stream_pd(a+j+2,_mm_add_pd(_mm_load_pd(b+j+2),_mm_mul_pd(scalar2,_mm_load_pd(c+j+2))));
237         _mm_stream_pd(a+j+4,_mm_add_pd(_mm_load_pd(b+j+4),_mm_mul_pd(scalar2,_mm_load_pd(c+j+4))));
238         _mm_stream_pd(a+j+6,_mm_add_pd(_mm_load_pd(b+j+6),_mm_mul_pd(scalar2,_mm_load_pd(c+j+6))));
239 #  if PREFETCH_NTA
240         _mm_prefetch(b+j+64,_MM_HINT_NTA);
241         _mm_prefetch(c+j+64,_MM_HINT_NTA);
242 #  endif
243       }
244     }
245 #else
246     for (j=0; j<N; j++) a[j] = b[j]+scalar*c[j];
247 #endif
248     MPI_Barrier(PETSC_COMM_WORLD);
249     times[3][k] = Second() - times[3][k];
250   }
251 
252   /*	--- SUMMARY --- */
253 
254   for (k=0; k<NTIMES; k++) {
255     for (j=0; j<4; j++) {
256       rmstime[j] = rmstime[j] + (times[j][k] * times[j][k]);
257       mintime[j] = MIN(mintime[j], times[j][k]);
258       maxtime[j] = MAX(maxtime[j], times[j][k]);
259     }
260   }
261 
262 
263   PetscPrintf(PETSC_COMM_WORLD,"%8s:  %11s  %11s  %11s  %11s  %11s\n","Function","Rate (MB/s)","Total (MB/s)","RMS time","Min time","Max time");
264   for (j=0; j<4; j++) {
265     rmstime[j] = sqrt(rmstime[j]/(double)NTIMES);
266     PetscPrintf(PETSC_COMM_WORLD,"%8s: %11.4f  %11.4f  %11.4f  %11.4f  %11.4f\n", label[j], 1.0e-06*bytes[j]/mintime[j], size*1.0e-06*bytes[j]/mintime[j], rmstime[j], mintime[j], maxtime[j]);
267   }
268   PetscFinalize();
269   return 0;
270 }
271 
272 static double Second() {
273   double t;
274   PetscTime(t);
275   return t;
276 }
277 
278 #define M 20
279 static int checktick()
280 {
281   int		i, minDelta, Delta;
282   double	t1, t2, timesfound[M];
283 
284   /*  Collect a sequence of M unique time values from the system. */
285 
286   for (i = 0; i < M; i++) {
287     t1 = Second();
288     while ((t2 = Second()) - t1 < 1.0E-6) {}
289     timesfound[i] = t1 = t2;
290   }
291 
292   /*
293    * Determine the minimum difference between these M values.
294    * This result will be our estimate (in microseconds) for the
295    * clock granularity.
296    */
297 
298   minDelta = 1000000;
299   for (i = 1; i < M; i++) {
300     Delta = (int)( 1.0E6 * (timesfound[i]-timesfound[i-1]));
301     minDelta = MIN(minDelta, MAX(Delta,0));
302   }
303 
304   return(minDelta);
305 }
306