16 …rror SSE2 instruction set is not enabled, try adding -march=native to CFLAGS or disable by adding …
18 #if !defined(PREFETCH_NTA) /* Use software prefetch and set non-temporal policy so that lines evict…
34  * Revision: 4.0-BETA, October 24, 1995
40  * the cost of floating-point operations relative to memory accesses.
46  *          at least 20 clock-ticks.  This will provide rate estimates
54 #define HLINE "-------------------------------------------------------------\n"
76   int          BytesPerWord, j, k, size;  in main()  local
77   PetscInt     node = -1;  in main()
85   PetscCall(PetscOptionsGetInt(NULL, NULL, "-node", &node, NULL));  in main()
86   /* --- SETUP --- determine precision and check timing --- */  in main()
101   if (node == -1) {  in main()
105   } else if (node == -2) {  in main()
118   for (j = 0; j < N; j++) {  in main()
119     a[j] = 1.0;  in main()
120     b[j] = 2.0;  in main()
121     c[j] = 0.0;  in main()
124   for (j = 0; j < N; j++) a[j] = 1.0;  in main()
125   for (j = 0; j < N; j++) b[j] = 2.0;  in main()
126   for (j = 0; j < N; j++) c[j] = 0.0;  in main()
135   for (j = 0; j < N; j++) a[j] = 2.0E0 * a[j];  in main()
136   t = 1.0E6 * (Second() - t);  in main()
145   PetscPrintf(PETSC_COMM_WORLD, "WARNING -- The above is only a rough guideline.\n");  in main()
150   /* --- MAIN LOOP --- repeat test cases NTIMES times --- */  in main()
155     /* ### COPY: c <- a ### */  in main()
161     for (j = 0; j < N; j += 8) {  in main()
162       _mm_stream_pd(c + j + 0, _mm_load_pd(a + j + 0));  in main()
163       _mm_stream_pd(c + j + 2, _mm_load_pd(a + j + 2));  in main()
164       _mm_stream_pd(c + j + 4, _mm_load_pd(a + j + 4));  in main()
165       _mm_stream_pd(c + j + 6, _mm_load_pd(a + j + 6));  in main()
167       _mm_prefetch(a + j + 64, _MM_HINT_NTA);  in main()
171     for (j = 0; j < N; j++) c[j] = a[j];  in main()
174     times[0][k] = Second() - times[0][k];  in main()
176     /* ### SCALE: b <- scalar * c ### */  in main()
182       for (j = 0; j < N; j += 8) {  in main()
183         _mm_stream_pd(b + j + 0, _mm_mul_pd(scalar2, _mm_load_pd(c + j + 0)));  in main()
184         _mm_stream_pd(b + j + 2, _mm_mul_pd(scalar2, _mm_load_pd(c + j + 2)));  in main()
185         _mm_stream_pd(b + j + 4, _mm_mul_pd(scalar2, _mm_load_pd(c + j + 4)));  in main()
186         _mm_stream_pd(b + j + 6, _mm_mul_pd(scalar2, _mm_load_pd(c + j + 6)));  in main()
188         _mm_prefetch(c + j + 64, _MM_HINT_NTA);  in main()
193     for (j = 0; j < N; j++) b[j] = scalar * c[j];  in main()
196     times[1][k] = Second() - times[1][k];  in main()
198     /* ### ADD: c <- a + b ### */  in main()
203       for (j = 0; j < N; j += 8) {  in main()
204         _mm_stream_pd(c + j + 0, _mm_add_pd(_mm_load_pd(a + j + 0), _mm_load_pd(b + j + 0)));  in main()
205         _mm_stream_pd(c + j + 2, _mm_add_pd(_mm_load_pd(a + j + 2), _mm_load_pd(b + j + 2)));  in main()
206         _mm_stream_pd(c + j + 4, _mm_add_pd(_mm_load_pd(a + j + 4), _mm_load_pd(b + j + 4)));  in main()
207         _mm_stream_pd(c + j + 6, _mm_add_pd(_mm_load_pd(a + j + 6), _mm_load_pd(b + j + 6)));  in main()
209         _mm_prefetch(a + j + 64, _MM_HINT_NTA);  in main()
210         _mm_prefetch(b + j + 64, _MM_HINT_NTA);  in main()
215     for (j = 0; j < N; j++) c[j] = a[j] + b[j];  in main()
218     times[2][k] = Second() - times[2][k];  in main()
220     /* ### TRIAD: a <- b + scalar * c ### */  in main()
226       for (j = 0; j < N; j += 8) {  in main()
227 …_mm_stream_pd(a + j + 0, _mm_add_pd(_mm_load_pd(b + j + 0), _mm_mul_pd(scalar2, _mm_load_pd(c + j …  in main()
228 …_mm_stream_pd(a + j + 2, _mm_add_pd(_mm_load_pd(b + j + 2), _mm_mul_pd(scalar2, _mm_load_pd(c + j …  in main()
229 …_mm_stream_pd(a + j + 4, _mm_add_pd(_mm_load_pd(b + j + 4), _mm_mul_pd(scalar2, _mm_load_pd(c + j …  in main()
230 …_mm_stream_pd(a + j + 6, _mm_add_pd(_mm_load_pd(b + j + 6), _mm_mul_pd(scalar2, _mm_load_pd(c + j …  in main()
232         _mm_prefetch(b + j + 64, _MM_HINT_NTA);  in main()
233         _mm_prefetch(c + j + 64, _MM_HINT_NTA);  in main()
238     for (j = 0; j < N; j++) a[j] = b[j] + scalar * c[j];  in main()
241     times[3][k] = Second() - times[3][k];  in main()
244   /* --- SUMMARY --- */  in main()
247     for (j = 0; j < 4; j++) {  in main()
248       rmstime[j] = rmstime[j] + (times[j][k] * times[j][k]);  in main()
249       mintime[j] = MIN(mintime[j], times[j][k]);  in main()
250       maxtime[j] = MAX(maxtime[j], times[j][k]);  in main()
254   for (j = 0; j < 4; j++) {  in main()
255     rmstime[j] = sqrt(rmstime[j] / (double)NTIMES);  in main()
256 …4f  %11.4f\n", label[j], 1.0e-06 * bytes[j] / mintime[j], size * 1.0e-06 * bytes[j] / mintime[j], …  in main()
279     while ((t2 = Second()) - t1 < 1.0E-6) { }  in checktick()
291     Delta    = (int)(1.0E6 * (timesfound[i] - timesfound[i - 1]));  in checktick()