Révision 143
Pi/C/Hybrid/Pi_aHybrid.c (revision 143) | ||
---|---|---|
1 |
// |
|
2 |
// Estimation of Pi using Monte Carlo exploration process |
|
3 |
// gcc -std=c99 -O3 -o Pi Pi.c -lm |
|
4 |
// Emmanuel Quemener <emmanuel.quemener@ens-lyon.fr> |
|
5 |
// Cecill v2 |
|
6 |
|
|
7 |
// Needed for gethostname |
|
8 |
#define _BSD_SOURCE |
|
9 |
#include <sys/unistd.h> |
|
10 |
|
|
11 |
#include <math.h> |
|
12 |
#include <stdio.h> |
|
13 |
#include <stdlib.h> |
|
14 |
#include <limits.h> |
|
15 |
#include <mpi.h> |
|
16 |
#include <stddef.h> |
|
17 |
|
|
18 |
#ifdef TIME |
|
19 |
#include <sys/time.h> |
|
20 |
#endif |
|
21 |
|
|
22 |
// Marsaglia RNG very simple implementation |
|
23 |
#define znew ((z=36969*(z&65535)+(z>>16))<<16) |
|
24 |
#define wnew ((w=18000*(w&65535)+(w>>16))&65535) |
|
25 |
#define MWC (znew+wnew) |
|
26 |
#define SHR3 (jsr=(jsr=(jsr=jsr^(jsr<<17))^(jsr>>13))^(jsr<<5)) |
|
27 |
#define CONG (jcong=69069*jcong+1234567) |
|
28 |
#define KISS ((MWC^CONG)+SHR3) |
|
29 |
|
|
30 |
#define ITERATIONS 1000000000 |
|
31 |
|
|
32 |
#define MWCfp MWC * 2.328306435454494e-10f |
|
33 |
#define KISSfp KISS * 2.328306435454494e-10f |
|
34 |
#define SHR3fp SHR3 * 2.328306435454494e-10f |
|
35 |
#define CONGfp CONG * 2.328306435454494e-10f |
|
36 |
|
|
37 |
#define PROCESS 1 |
|
38 |
|
|
39 |
#ifdef LONG |
|
40 |
#define LENGTH long long |
|
41 |
#else |
|
42 |
#define LENGTH int |
|
43 |
#endif |
|
44 |
|
|
45 |
typedef struct compute_node { |
|
46 |
LENGTH iterations; |
|
47 |
int process; |
|
48 |
} node; |
|
49 |
|
|
50 |
typedef struct compute_result { |
|
51 |
LENGTH inside; |
|
52 |
long int useconds; |
|
53 |
} result; |
|
54 |
|
|
55 |
unsigned int rotl(unsigned int value, int shift) { |
|
56 |
return (value << shift) | (value >> (sizeof(value) * CHAR_BIT - shift)); |
|
57 |
} |
|
58 |
|
|
59 |
unsigned int rotr(unsigned int value, int shift) { |
|
60 |
return (value >> shift) | (value << (sizeof(value) * CHAR_BIT - shift)); |
|
61 |
} |
|
62 |
|
|
63 |
LENGTH MainLoopGlobal(LENGTH iterations,unsigned int seed_w,unsigned int seed_z) |
|
64 |
{ |
|
65 |
|
|
66 |
#if defined TCONG |
|
67 |
unsigned int jcong=seed_z; |
|
68 |
#elif defined TSHR3 |
|
69 |
unsigned int jsr=seed_w; |
|
70 |
#elif defined TMWC |
|
71 |
unsigned int z=seed_z; |
|
72 |
unsigned int w=seed_w; |
|
73 |
#elif defined TKISS |
|
74 |
unsigned int jcong=seed_z; |
|
75 |
unsigned int jsr=seed_w; |
|
76 |
unsigned int z=seed_z; |
|
77 |
unsigned int w=seed_w; |
|
78 |
#endif |
|
79 |
|
|
80 |
LENGTH total=0; |
|
81 |
|
|
82 |
for (LENGTH i=0;i<iterations;i++) { |
|
83 |
|
|
84 |
#if defined TINT32 |
|
85 |
#define THEONE 1073741824 |
|
86 |
#if defined TCONG |
|
87 |
unsigned int x=CONG>>17 ; |
|
88 |
unsigned int y=CONG>>17 ; |
|
89 |
#elif defined TSHR3 |
|
90 |
unsigned int x=SHR3>>17 ; |
|
91 |
unsigned int y=SHR3>>17 ; |
|
92 |
#elif defined TMWC |
|
93 |
unsigned int x=MWC>>17 ; |
|
94 |
unsigned int y=MWC>>17 ; |
|
95 |
#elif defined TKISS |
|
96 |
unsigned int x=KISS>>17 ; |
|
97 |
unsigned int y=KISS>>17 ; |
|
98 |
#endif |
|
99 |
#elif defined TINT64 |
|
100 |
#define THEONE 4611686018427387904 |
|
101 |
#if defined TCONG |
|
102 |
unsigned long x=(unsigned long)(CONG>>1) ; |
|
103 |
unsigned long y=(unsigned long)(CONG>>1) ; |
|
104 |
#elif defined TSHR3 |
|
105 |
unsigned long x=(unsigned long)(SHR3>>1) ; |
|
106 |
unsigned long y=(unsigned long)(SHR3>>1) ; |
|
107 |
#elif defined TMWC |
|
108 |
unsigned long x=(unsigned long)(MWC>>1) ; |
|
109 |
unsigned long y=(unsigned long)(MWC>>1) ; |
|
110 |
#elif defined TKISS |
|
111 |
unsigned long x=(unsigned long)(KISS>>1) ; |
|
112 |
unsigned long y=(unsigned long)(KISS>>1) ; |
|
113 |
#endif |
|
114 |
#elif defined TFP32 |
|
115 |
#define THEONE 1.0f |
|
116 |
#if defined TCONG |
|
117 |
float x=CONGfp ; |
|
118 |
float y=CONGfp ; |
|
119 |
#elif defined TSHR3 |
|
120 |
float x=SHR3fp ; |
|
121 |
float y=SHR3fp ; |
|
122 |
#elif defined TMWC |
|
123 |
float x=MWCfp ; |
|
124 |
float y=MWCfp ; |
|
125 |
#elif defined TKISS |
|
126 |
float x=KISSfp ; |
|
127 |
float y=KISSfp ; |
|
128 |
#endif |
|
129 |
#elif defined TFP64 |
|
130 |
#define THEONE 1.0f |
|
131 |
#if defined TCONG |
|
132 |
double x=(double)CONGfp ; |
|
133 |
double y=(double)CONGfp ; |
|
134 |
#elif defined TSHR3 |
|
135 |
double x=(double)SHR3fp ; |
|
136 |
double y=(double)SHR3fp ; |
|
137 |
#elif defined TMWC |
|
138 |
double x=(double)MWCfp ; |
|
139 |
double y=(double)MWCfp ; |
|
140 |
#elif defined TKISS |
|
141 |
double x=(double)KISSfp ; |
|
142 |
double y=(double)KISSfp ; |
|
143 |
#endif |
|
144 |
#endif |
|
145 |
|
|
146 |
// Matching test |
|
147 |
unsigned long inside=((x*x+y*y) < THEONE) ? 1:0; |
|
148 |
total+=inside; |
|
149 |
} |
|
150 |
|
|
151 |
return(total); |
|
152 |
} |
|
153 |
|
|
154 |
int main(int argc, char *argv[]) { |
|
155 |
|
|
156 |
unsigned int seed_z=362436069,seed_w=52128862,process=PROCESS; |
|
157 |
// Number of NP or OpenMP processes <1024 |
|
158 |
LENGTH iterations=ITERATIONS,insideMPI[8192],insideOpenMP[1024], |
|
159 |
part_inside=0,part_iterations,insides=0; |
|
160 |
int numtasks,rank,rc,tag=1,i; |
|
161 |
float pi; |
|
162 |
|
|
163 |
// Hostname supposed to be <128 characters |
|
164 |
char hostname[128]; |
|
165 |
|
|
166 |
gethostname(hostname, sizeof hostname); |
|
167 |
|
|
168 |
struct timeval start,end; |
|
169 |
long int useconds; |
|
170 |
|
|
171 |
MPI_Status Stat; |
|
172 |
MPI_Request RequestSend,RequestRecv,RequestSend2,RequestRecv2; |
|
173 |
|
|
174 |
rc = MPI_Init(&argc,&argv); |
|
175 |
if (rc != MPI_SUCCESS) { |
|
176 |
printf ("Error starting MPI program. Terminating.\n"); |
|
177 |
MPI_Abort(MPI_COMM_WORLD, rc); |
|
178 |
} |
|
179 |
|
|
180 |
MPI_Comm_size(MPI_COMM_WORLD,&numtasks); |
|
181 |
MPI_Comm_rank(MPI_COMM_WORLD,&rank); |
|
182 |
|
|
183 |
const int nitems=2; |
|
184 |
int blocklengths[2] = {1,1}; |
|
185 |
|
|
186 |
#ifdef LONG |
|
187 |
MPI_Datatype types_node[2] = {MPI_LONG, MPI_INT}; |
|
188 |
MPI_Datatype types_result[2] = {MPI_LONG, MPI_LONG}; |
|
189 |
#else |
|
190 |
MPI_Datatype types_node[2] = {MPI_INT, MPI_INT}; |
|
191 |
MPI_Datatype types_result[2] = {MPI_INT, MPI_LONG}; |
|
192 |
#endif |
|
193 |
|
|
194 |
MPI_Datatype mpi_node_type,mpi_result_type; |
|
195 |
MPI_Aint offsets[2],offsetsr[2]; |
|
196 |
|
|
197 |
offsets[0] = offsetof(node, iterations); |
|
198 |
offsets[1] = offsetof(node, process); |
|
199 |
|
|
200 |
MPI_Type_create_struct(nitems, blocklengths, offsets, types_node, &mpi_node_type); |
|
201 |
MPI_Type_commit(&mpi_node_type); |
|
202 |
|
|
203 |
offsetsr[0] = offsetof(result, inside); |
|
204 |
offsetsr[1] = offsetof(result, useconds); |
|
205 |
|
|
206 |
MPI_Type_create_struct(nitems, blocklengths, offsetsr, types_result, &mpi_result_type); |
|
207 |
MPI_Type_commit(&mpi_result_type); |
|
208 |
|
|
209 |
if (rank==0) { |
|
210 |
|
|
211 |
if (argc > 1) { |
|
212 |
iterations=(LENGTH)atoll(argv[1]); |
|
213 |
process=atoi(argv[2]); |
|
214 |
} |
|
215 |
else { |
|
216 |
printf("\n\tPi : Estimate Pi with Monte Carlo exploration\n\n"); |
|
217 |
printf("\t\t#1 : number of iterations (default 1 billion)\n\n"); |
|
218 |
printf("\t\t#1 : number of OpenMP processes (default 1)\n\n"); |
|
219 |
} |
|
220 |
|
|
221 |
printf ("\n\tInformation about architecture:\n\n"); |
|
222 |
|
|
223 |
printf ("\tSizeof int = %lld bytes.\n", (long long)sizeof(int)); |
|
224 |
printf ("\tSizeof long = %lld bytes.\n", (long long)sizeof(long)); |
|
225 |
printf ("\tSizeof long long = %lld bytes.\n", (long long)sizeof(long long)); |
|
226 |
|
|
227 |
printf ("\tMax int = %u\n", INT_MAX); |
|
228 |
printf ("\tMax long = %ld\n", LONG_MAX); |
|
229 |
printf ("\tMax long long = %lld\n\n", LLONG_MAX); |
|
230 |
|
|
231 |
part_iterations=(((iterations%numtasks)%process) == 0) ? iterations/numtasks/process:iterations/numtasks/process+1 ; |
|
232 |
|
|
233 |
node send; |
|
234 |
send.iterations=part_iterations; |
|
235 |
send.process=process; |
|
236 |
|
|
237 |
// Split part of code |
|
238 |
for (i=1;i<numtasks;i++) { |
|
239 |
rc = MPI_Isend(&send, 1, mpi_node_type, i, tag, MPI_COMM_WORLD, &RequestSend); |
|
240 |
} |
|
241 |
MPI_Wait(&RequestSend,&Stat); |
|
242 |
|
|
243 |
gettimeofday(&start,(struct timezone *)0); |
|
244 |
|
|
245 |
#pragma omp parallel for |
|
246 |
for (int i=0 ; i<process; i++) { |
|
247 |
insideOpenMP[i]=MainLoopGlobal(part_iterations, |
|
248 |
rotr(seed_w,i), |
|
249 |
rotl(seed_z,i)); |
|
250 |
/* |
|
251 |
printf("\t(%s,%i) found %lld for process %i\n",hostname,0, |
|
252 |
(long long)insideOpenMP[i],i); */ |
|
253 |
} |
|
254 |
/* printf("\n"); */ |
|
255 |
|
|
256 |
for (int i=0 ; i<process; i++) { |
|
257 |
insides+=insideOpenMP[i]; |
|
258 |
} |
|
259 |
|
|
260 |
gettimeofday(&end,(struct timezone *)0); |
|
261 |
useconds=(end.tv_sec-start.tv_sec)*1000000+end.tv_usec-start.tv_usec; |
|
262 |
|
|
263 |
printf("\tOn %s with rank #%i find %lld inside in %lu useconds.\n", |
|
264 |
hostname,rank,(long long)insides,useconds); |
|
265 |
|
|
266 |
// Join part of code |
|
267 |
for (i=1;i<numtasks;i++) { |
|
268 |
|
|
269 |
result recv; |
|
270 |
|
|
271 |
rc = MPI_Irecv(&recv,1,mpi_result_type,i,tag,MPI_COMM_WORLD,&RequestRecv2); |
|
272 |
MPI_Wait(&RequestRecv2,&Stat); |
|
273 |
|
|
274 |
insideMPI[i]=recv.inside; |
|
275 |
useconds=recv.useconds; |
|
276 |
|
|
277 |
printf("\tReceive from rank #%i, find %lld inside in %lu useconds\n",i,(long long)insideMPI[i],useconds); |
|
278 |
|
|
279 |
insides+=insideMPI[i]; |
|
280 |
} |
|
281 |
|
|
282 |
pi=4.*(float)insides/(float)(part_iterations*numtasks*process); |
|
283 |
|
|
284 |
printf("\n\tPi=%.40f\n\twith error %.40f\n\twith %lld iterations\n\n",pi, |
|
285 |
fabs(pi-4*atan(1.))/pi,(long long)(part_iterations*numtasks*process)); |
|
286 |
|
|
287 |
} |
|
288 |
else |
|
289 |
{ |
|
290 |
// Receive information from master |
|
291 |
|
|
292 |
node recv; |
|
293 |
|
|
294 |
rc = MPI_Irecv(&recv,1,mpi_node_type,0,tag,MPI_COMM_WORLD,&RequestRecv); |
|
295 |
MPI_Wait(&RequestRecv,&Stat); |
|
296 |
/* |
|
297 |
printf("\t(%s,%i) receive from master %lld with %i process\n", |
|
298 |
hostname,rank,(long long)recv.iterations,recv.process); |
|
299 |
*/ |
|
300 |
|
|
301 |
gettimeofday(&start,(struct timezone *)0); |
|
302 |
|
|
303 |
#pragma omp parallel for |
|
304 |
for (int i=0 ; i<recv.process; i++) { |
|
305 |
insideOpenMP[i]=MainLoopGlobal(recv.iterations,rotr(seed_w,rank+i),rotl(seed_z,rank-i)); |
|
306 |
/* |
|
307 |
printf("\t(%s,%i) found %lld for process %i\n",hostname,rank, |
|
308 |
(long long)insideOpenMP[i],i); |
|
309 |
*/ |
|
310 |
} |
|
311 |
/* printf("\n"); */ |
|
312 |
|
|
313 |
for (int i=0 ; i<recv.process; i++) { |
|
314 |
part_inside+=insideOpenMP[i]; |
|
315 |
} |
|
316 |
|
|
317 |
gettimeofday(&end,(struct timezone *)0); |
|
318 |
useconds=(end.tv_sec-start.tv_sec)*1000000+end.tv_usec-start.tv_usec; |
|
319 |
|
|
320 |
/* |
|
321 |
printf("\tOn %s rank %i find %lld inside in %lu useconds.\n", |
|
322 |
hostname,rank,(long long)part_inside,useconds); |
|
323 |
*/ |
|
324 |
|
|
325 |
result send; |
|
326 |
send.inside=part_inside; |
|
327 |
send.useconds=useconds; |
|
328 |
|
|
329 |
rc = MPI_Isend(&send, 1, mpi_result_type, 0, tag, MPI_COMM_WORLD, &RequestSend2); |
|
330 |
|
|
331 |
MPI_Wait(&RequestSend2, &Stat); |
|
332 |
} |
|
333 |
|
|
334 |
MPI_Type_free(&mpi_node_type); |
|
335 |
MPI_Type_free(&mpi_result_type); |
|
336 |
|
|
337 |
MPI_Finalize(); |
|
338 |
} |
Formats disponibles : Unified diff