1 | #include "stdio.h" |
---|
2 | |
---|
3 | //////////////////////////////////// |
---|
4 | // Image parameters |
---|
5 | |
---|
6 | #define PIXEL_SIZE 2 |
---|
7 | #define NL 1024 |
---|
8 | #define NP 1024 |
---|
9 | #define BLOCK_SIZE 1024 |
---|
10 | |
---|
11 | #define PRINTF if(lid==0) tty_printf |
---|
12 | |
---|
13 | #define TA(c,l,p) (A[c][((NP)*(l))+(p)]) |
---|
14 | #define TB(c,p,l) (B[c][((NL)*(p))+(l)]) |
---|
15 | #define TC(c,l,p) (C[c][((NP)*(l))+(p)]) |
---|
16 | #define TD(c,l,p) (D[c][((NP)*(l))+(p)]) |
---|
17 | #define TZ(c,l,p) (Z[c][((NP)*(l))+(p)]) |
---|
18 | |
---|
19 | #define max(x,y) ((x) > (y) ? (x) : (y)) |
---|
20 | #define min(x,y) ((x) < (y) ? (x) : (y)) |
---|
21 | |
---|
22 | /////////////////////////////////////////// |
---|
23 | // tricks to read parameters from ldscript |
---|
24 | /////////////////////////////////////////// |
---|
25 | |
---|
26 | struct plaf; |
---|
27 | |
---|
28 | extern struct plaf seg_heap_base; |
---|
29 | extern struct plaf NB_PROCS; |
---|
30 | extern struct plaf NB_CLUSTERS; |
---|
31 | |
---|
32 | ///////////// |
---|
33 | void main() |
---|
34 | { |
---|
35 | |
---|
36 | ////////////////////////////////// |
---|
37 | // convolution kernel parameters |
---|
38 | // The content of this section is |
---|
39 | // Philips proprietary information. |
---|
40 | /////////////////////////////////// |
---|
41 | |
---|
42 | int vnorm = 115; |
---|
43 | int vf[35]; |
---|
44 | vf[0] = 1; |
---|
45 | vf[1] = 1; |
---|
46 | vf[2] = 2; |
---|
47 | vf[3] = 2; |
---|
48 | vf[4] = 2; |
---|
49 | vf[5] = 2; |
---|
50 | vf[6] = 3; |
---|
51 | vf[7] = 3; |
---|
52 | vf[8] = 3; |
---|
53 | vf[9] = 4; |
---|
54 | vf[10] = 4; |
---|
55 | vf[11] = 4; |
---|
56 | vf[12] = 4; |
---|
57 | vf[13] = 5; |
---|
58 | vf[14] = 5; |
---|
59 | vf[15] = 5; |
---|
60 | vf[16] = 5; |
---|
61 | vf[17] = 5; |
---|
62 | vf[18] = 5; |
---|
63 | vf[19] = 5; |
---|
64 | vf[20] = 5; |
---|
65 | vf[21] = 5; |
---|
66 | vf[22] = 4; |
---|
67 | vf[23] = 4; |
---|
68 | vf[24] = 4; |
---|
69 | vf[25] = 4; |
---|
70 | vf[26] = 3; |
---|
71 | vf[27] = 3; |
---|
72 | vf[28] = 3; |
---|
73 | vf[29] = 2; |
---|
74 | vf[30] = 2; |
---|
75 | vf[31] = 2; |
---|
76 | vf[32] = 2; |
---|
77 | vf[33] = 1; |
---|
78 | vf[34] = 1; |
---|
79 | |
---|
80 | int hrange = 100; |
---|
81 | int hnorm = 201; |
---|
82 | |
---|
83 | unsigned int date = 0; |
---|
84 | |
---|
85 | int c; // cluster index for loops |
---|
86 | int l; // line index for loops |
---|
87 | int p; // pixel index for loops |
---|
88 | int x; // filter index for loops |
---|
89 | |
---|
90 | int pid = procid(); // processor id |
---|
91 | int nprocs = (int)&NB_PROCS; // number of processors per cluster |
---|
92 | int nclusters = (int)&NB_CLUSTERS; // number of clusters |
---|
93 | int lid = pid%nprocs; // local task id |
---|
94 | int cid = pid/nprocs; // cluster task id |
---|
95 | int base = (unsigned int)&seg_heap_base; // base address for shared buffers |
---|
96 | int increment = (0x80000000 / nclusters) * 2; // cluster increment |
---|
97 | int ntasks = nclusters * nprocs; // number of tasks |
---|
98 | int nblocks = (NP*NL*PIXEL_SIZE)/BLOCK_SIZE; // number of blocks per image |
---|
99 | |
---|
100 | int lines_per_task = NL/ntasks; // number of lines per task |
---|
101 | int lines_per_cluster = NL/nclusters; // number of lines per cluster |
---|
102 | int pixels_per_task = NP/ntasks; // number of columns per task |
---|
103 | int pixels_per_cluster = NP/nclusters; // number of columns per cluster |
---|
104 | |
---|
105 | int first, last; |
---|
106 | |
---|
107 | PRINTF("\n*** Processor %d entering main at cycle %d ***\n\n", pid, proctime()); |
---|
108 | |
---|
109 | ////////////////////////// |
---|
110 | // parameters checking |
---|
111 | if( (nprocs != 1) && (nprocs != 2) && (nprocs != 4) ) |
---|
112 | { |
---|
113 | PRINTF("NB_PROCS must be 1, 2 or 4\n"); |
---|
114 | while(1); |
---|
115 | } |
---|
116 | if( (nclusters != 4) && (nclusters != 8) && (nclusters != 16) && |
---|
117 | (nclusters != 32) && (nclusters != 64) && (nclusters !=128) && (nclusters != 256) ) |
---|
118 | { |
---|
119 | PRINTF("NB_CLUSTERS must be a power of 2 between 4 and 256\n"); |
---|
120 | while(1); |
---|
121 | } |
---|
122 | if( pid >= ntasks ) |
---|
123 | { |
---|
124 | PRINTF("processor id %d larger than NB_CLUSTERS*NB_PROCS\n", pid); |
---|
125 | while(1); |
---|
126 | } |
---|
127 | if ( NL % nclusters != 0 ) |
---|
128 | { |
---|
129 | PRINTF("NB_CLUSTERS must be a divider of NL"); |
---|
130 | while(1); |
---|
131 | } |
---|
132 | if( NP % nclusters != 0 ) |
---|
133 | { |
---|
134 | PRINTF("NB_CLUSTERS must be a divider of NP"); |
---|
135 | while(1); |
---|
136 | } |
---|
137 | |
---|
138 | ////////////////////////////////////////////////////////////////// |
---|
139 | // Arrays of pointers on the shared, distributed buffers |
---|
140 | // containing the images (sized for the worst case : 256 clusters) |
---|
141 | unsigned short* A[256]; |
---|
142 | int* B[256]; |
---|
143 | int* C[256]; |
---|
144 | int* D[256]; |
---|
145 | unsigned char* Z[256]; |
---|
146 | |
---|
147 | // Arrays of pointers on the instrumentation arrays |
---|
148 | // These arrays are indexed by the cluster index (sized for the worst case : 256 clusters) |
---|
149 | // each pointer points on the base adress of an array of 4 (NPROCS max) unsigned int |
---|
150 | unsigned int* LOAD_START[256]; |
---|
151 | unsigned int* LOAD_ENDED[256]; |
---|
152 | unsigned int* VERT_START[256]; |
---|
153 | unsigned int* VERT_ENDED[256]; |
---|
154 | unsigned int* HORI_START[256]; |
---|
155 | unsigned int* HORI_ENDED[256]; |
---|
156 | unsigned int* DISP_START[256]; |
---|
157 | unsigned int* DISP_ENDED[256]; |
---|
158 | |
---|
159 | // The shared, distributed buffers addresses are computed |
---|
160 | // from the seg_heap_base value defined in the ldscript file |
---|
161 | // and from the cluster increment = 4Gbytes/nclusters. |
---|
162 | // These arrays of pointers are identical and |
---|
163 | // replicated in the stack of each task |
---|
164 | for( c=0 ; c<nclusters ; c++) |
---|
165 | { |
---|
166 | A[c] = (unsigned short*) (base + increment*c); |
---|
167 | Z[c] = (unsigned char*) (base + 2*NP*NL/nclusters + increment*c); |
---|
168 | B[c] = (int*) (base + 4*NP*NL/nclusters + increment*c); |
---|
169 | C[c] = (int*) (base + 8*NP*NL/nclusters + increment*c); |
---|
170 | D[c] = (int*) (base + 12*NP*NL/nclusters + increment*c); |
---|
171 | |
---|
172 | LOAD_START[c] = (unsigned int*) (base + 3*NL*NP/nclusters + increment*c); |
---|
173 | LOAD_ENDED[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 16 + increment*c); |
---|
174 | VERT_START[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 32 + increment*c); |
---|
175 | VERT_ENDED[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 48 + increment*c); |
---|
176 | HORI_START[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 64 + increment*c); |
---|
177 | HORI_ENDED[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 80 + increment*c); |
---|
178 | DISP_START[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 96 + increment*c); |
---|
179 | DISP_ENDED[c] = (unsigned int*) (base + 3*NL*NP/nclusters + 112 + increment*c); |
---|
180 | } |
---|
181 | |
---|
182 | PRINTF("NCLUSTERS = %d\n", nclusters); |
---|
183 | PRINTF("NPROCS = %d\n\n", nprocs); |
---|
184 | |
---|
185 | PRINTF("*** Starting barrier init at cycle %d ***\n", proctime()); |
---|
186 | |
---|
187 | // barriers initialization |
---|
188 | barrier_init(0, ntasks); |
---|
189 | barrier_init(1, ntasks); |
---|
190 | barrier_init(2, ntasks); |
---|
191 | barrier_init(3, ntasks); |
---|
192 | |
---|
193 | PRINTF("*** Completing barrier init at cycle %d ***\n", proctime()); |
---|
194 | |
---|
195 | //////////////////////////////////////////////////////// |
---|
196 | // pseudo parallel load from disk to A[c] buffers |
---|
197 | // only task running on processor with (lid==0) does it |
---|
198 | // nblocks/nclusters are loaded in each cluster |
---|
199 | |
---|
200 | if ( lid == 0 ) |
---|
201 | { |
---|
202 | int p; |
---|
203 | date = proctime(); |
---|
204 | PRINTF("\n*** Starting load at cycle %d\n", date); |
---|
205 | for ( p=0 ; p<nprocs ; p++ ) LOAD_START[cid][p] = date; |
---|
206 | |
---|
207 | if( ioc_read(nblocks*cid/nclusters, |
---|
208 | A[cid] , |
---|
209 | nblocks/nclusters) ) |
---|
210 | { |
---|
211 | PRINTF("echec ioc_read\n"); |
---|
212 | while(1); |
---|
213 | } |
---|
214 | if ( ioc_completed() ) |
---|
215 | { |
---|
216 | PRINTF("echec ioc_completed\n"); |
---|
217 | while(1); |
---|
218 | } |
---|
219 | |
---|
220 | date = proctime(); |
---|
221 | PRINTF("*** Completing load at cycle %d\n", date); |
---|
222 | for ( p=0 ; p<nprocs ; p++ ) LOAD_ENDED[cid][p] = date; |
---|
223 | } |
---|
224 | |
---|
225 | barrier_wait(0); |
---|
226 | |
---|
227 | ////////////////////////////////////////////////////////// |
---|
228 | // parallel horizontal filter : |
---|
229 | // B <= transpose(FH(A)) |
---|
230 | // D <= A - FH(A) |
---|
231 | // Each task computes (NL/ntasks) lines |
---|
232 | // The image must be extended : |
---|
233 | // if (z<0) TA(cid,l,z) == TA(cid,l,0) |
---|
234 | // if (z>NP-1) TA(cid,l,z) == TA(cid,l,NP-1) |
---|
235 | |
---|
236 | date = proctime(); |
---|
237 | PRINTF("\n*** Starting horizontal filter at cycle %d\n", date); |
---|
238 | HORI_START[cid][lid] = date; |
---|
239 | |
---|
240 | // l = absolute line index / p = absolute pixel index |
---|
241 | // first & last define which lines are handled by a given task(cid,lid) |
---|
242 | |
---|
243 | first = (cid*nprocs + lid)*lines_per_task; |
---|
244 | last = first + lines_per_task; |
---|
245 | |
---|
246 | for ( l=first ; l<last ; l++) |
---|
247 | { |
---|
248 | // src_c and src_l are the cluster index and the line index for A & D |
---|
249 | int src_c = l/lines_per_cluster; |
---|
250 | int src_l = l%lines_per_cluster; |
---|
251 | |
---|
252 | // We use the spécific values of the horizontal ep-filter for optimisation: |
---|
253 | // sum(p) = sum(p-1) + TA[p+hrange] - TA[p-hrange-1] |
---|
254 | // To minimize the number of tests, the loop on pixels is split in three domains |
---|
255 | |
---|
256 | int sum_p = (hrange+2)*TA(src_c, src_l, 0); |
---|
257 | for ( x = 1 ; x < hrange ; x++) sum_p = sum_p + TA(src_c, src_l, x); |
---|
258 | |
---|
259 | // first domain : from 0 to hrange |
---|
260 | for ( p=0 ; p<hrange+1 ; p++) |
---|
261 | { |
---|
262 | // dst_c and dst_p are the cluster index and the pixel index for B |
---|
263 | int dst_c = p/pixels_per_cluster; |
---|
264 | int dst_p = p%pixels_per_cluster; |
---|
265 | sum_p = sum_p + (int)TA(src_c, src_l, p+hrange) - (int)TA(src_c, src_l, 0); |
---|
266 | TB(dst_c, dst_p, l) = sum_p/hnorm; |
---|
267 | TD(src_c, src_l, p) = (int)TA(src_c, src_l, p) - sum_p/hnorm; |
---|
268 | } |
---|
269 | // second domain : from (hrange+1) to (NP-hrange-1) |
---|
270 | for ( p = hrange+1 ; p < NP-hrange ; p++) |
---|
271 | { |
---|
272 | // dst_c and dst_p are the cluster index and the pixel index for B |
---|
273 | int dst_c = p/pixels_per_cluster; |
---|
274 | int dst_p = p%pixels_per_cluster; |
---|
275 | sum_p = sum_p + (int)TA(src_c, src_l, p+hrange) - (int)TA(src_c, src_l, p-hrange-1); |
---|
276 | TB(dst_c, dst_p, l) = sum_p/hnorm; |
---|
277 | TD(src_c, src_l, p) = (int)TA(src_c, src_l, p) - sum_p/hnorm; |
---|
278 | } |
---|
279 | // third domain : from (NP-hrange) to (NP-1) |
---|
280 | for ( p = NP-hrange ; p < NP ; p++) |
---|
281 | { |
---|
282 | // dst_c and dst_p are the cluster index and the pixel index for B |
---|
283 | int dst_c = p/pixels_per_cluster; |
---|
284 | int dst_p = p%pixels_per_cluster; |
---|
285 | sum_p = sum_p + (int)TA(src_c, src_l, NP-1) - (int)TA(src_c, src_l, p-hrange-1); |
---|
286 | TB(dst_c, dst_p, l) = sum_p/hnorm; |
---|
287 | TD(src_c, src_l, p) = (int)TA(src_c, src_l, p) - sum_p/hnorm; |
---|
288 | } |
---|
289 | |
---|
290 | PRINTF(" - line %d computed at cycle %d\n", l, proctime()); |
---|
291 | } |
---|
292 | |
---|
293 | date = proctime(); |
---|
294 | PRINTF("*** Completing horizontal filter at cycle %d\n", date); |
---|
295 | HORI_ENDED[cid][lid] = date; |
---|
296 | |
---|
297 | barrier_wait(1); |
---|
298 | |
---|
299 | ////////////////////////////////////////////////////////// |
---|
300 | // parallel vertical filter : |
---|
301 | // C <= transpose(FV(B)) |
---|
302 | // Each task computes (NP/ntasks) columns |
---|
303 | // The image must be extended : |
---|
304 | // if (l<0) TB(cid,p,x) == TB(cid,p,0) |
---|
305 | // if (l>NL-1) TB(cid,p,x) == TB(cid,p,NL-1) |
---|
306 | |
---|
307 | date = proctime(); |
---|
308 | PRINTF("\n*** starting vertical filter at cycle %d\n", date); |
---|
309 | VERT_START[cid][lid] = date; |
---|
310 | |
---|
311 | // l = absolute line index / p = absolute pixel index |
---|
312 | // first & last define which pixels are handled by a given task(cid,lid) |
---|
313 | |
---|
314 | first = (cid*nprocs + lid)*pixels_per_task; |
---|
315 | last = first + pixels_per_task; |
---|
316 | |
---|
317 | for ( p=first ; p<last ; p++) |
---|
318 | { |
---|
319 | // src_c and src_p are the cluster index and the pixel index for B |
---|
320 | int src_c = p/pixels_per_cluster; |
---|
321 | int src_p = p%pixels_per_cluster; |
---|
322 | |
---|
323 | int sum_l; |
---|
324 | |
---|
325 | // We use the specific values of the vertical ep-filter |
---|
326 | // To minimize the number of tests, the NL lines are split in three domains |
---|
327 | |
---|
328 | // first domain : explicit computation for the first 18 values |
---|
329 | for ( l=0 ; l<18 ; l++) |
---|
330 | { |
---|
331 | // dst_c and dst_l are the cluster index and the line index for C |
---|
332 | int dst_c = l/lines_per_cluster; |
---|
333 | int dst_l = l%lines_per_cluster; |
---|
334 | |
---|
335 | for ( x=0, sum_l=0 ; x<35 ; x++ ) |
---|
336 | { |
---|
337 | sum_l = sum_l + vf[x] * TB(src_c, src_p, max(l-17+x,0) ); |
---|
338 | } |
---|
339 | TC(dst_c, dst_l, p) = sum_l/vnorm; |
---|
340 | } |
---|
341 | // second domain |
---|
342 | for ( l = 18 ; l < NL-17 ; l++ ) |
---|
343 | { |
---|
344 | // dst_c and dst_l are the cluster index and the line index for C |
---|
345 | int dst_c = l/lines_per_cluster; |
---|
346 | int dst_l = l%lines_per_cluster; |
---|
347 | |
---|
348 | sum_l = sum_l + TB(src_c, src_p, l+4) |
---|
349 | + TB(src_c, src_p, l+8) |
---|
350 | + TB(src_c, src_p, l+11) |
---|
351 | + TB(src_c, src_p, l+15) |
---|
352 | + TB(src_c, src_p, l+17) |
---|
353 | - TB(src_c, src_p, l-5) |
---|
354 | - TB(src_c, src_p, l-9) |
---|
355 | - TB(src_c, src_p, l-12) |
---|
356 | - TB(src_c, src_p, l-16) |
---|
357 | - TB(src_c, src_p, l-18); |
---|
358 | TC(dst_c, dst_l, p) = sum_l/vnorm; |
---|
359 | } |
---|
360 | // third domain |
---|
361 | for ( l = NL-17 ; l < NL ; l++ ) |
---|
362 | { |
---|
363 | // dst_c and dst_l are the cluster index and the line index for C |
---|
364 | int dst_c = l/lines_per_cluster; |
---|
365 | int dst_l = l%lines_per_cluster; |
---|
366 | |
---|
367 | sum_l = sum_l + TB(src_c, src_p, min(l+4,NL-1)) |
---|
368 | + TB(src_c, src_p, min(l+8,NL-1)) |
---|
369 | + TB(src_c, src_p, min(l+11,NL-1)) |
---|
370 | + TB(src_c, src_p, min(l+15,NL-1)) |
---|
371 | + TB(src_c, src_p, min(l+17,NL-1)) |
---|
372 | - TB(src_c, src_p, l-5) |
---|
373 | - TB(src_c, src_p, l-9) |
---|
374 | - TB(src_c, src_p, l-12) |
---|
375 | - TB(src_c, src_p, l-16) |
---|
376 | - TB(src_c, src_p, l-18); |
---|
377 | TC(dst_c, dst_l, p) = sum_l/vnorm; |
---|
378 | } |
---|
379 | PRINTF(" - column %d computed at cycle %d\n", p, proctime()); |
---|
380 | } |
---|
381 | |
---|
382 | date = proctime(); |
---|
383 | PRINTF("*** Completing vertical filter at cycle %d\n", date); |
---|
384 | VERT_ENDED[cid][lid] = date; |
---|
385 | |
---|
386 | barrier_wait(2); |
---|
387 | |
---|
388 | //////////////////////////////////////////////////////////////// |
---|
389 | // final computation and parallel display |
---|
390 | // Z <= D + C |
---|
391 | // Each processor use its private DMA channel to display |
---|
392 | // the resulting image, line per line (one byte per pixel). |
---|
393 | // Eah processor computes & displays (NL/ntasks) lines. |
---|
394 | |
---|
395 | date = proctime(); |
---|
396 | PRINTF("\n*** Starting display at cycle %d\n", date); |
---|
397 | DISP_START[cid][lid] = date; |
---|
398 | |
---|
399 | first = lid*lines_per_task; |
---|
400 | last = first + lines_per_task; |
---|
401 | |
---|
402 | for ( l=first ; l<last ; l++) |
---|
403 | { |
---|
404 | for ( p=0 ; p<NP ; p++) |
---|
405 | { |
---|
406 | TZ(cid,l,p) = (unsigned char)(((TD(cid,l,p) + TC(cid,l,p))>>8) & 0xFF); |
---|
407 | } |
---|
408 | fb_write(NP*(cid*lines_per_cluster+l), &TZ(cid,l,0), NP); |
---|
409 | } |
---|
410 | |
---|
411 | date = proctime(); |
---|
412 | PRINTF("*** Completing display at cycle %d\n", date); |
---|
413 | DISP_ENDED[cid][lid] = date; |
---|
414 | |
---|
415 | barrier_wait(3); |
---|
416 | |
---|
417 | ///////////////////////////////////////////////////////// |
---|
418 | // Instrumentation (done by processor 0 in cluster 0) |
---|
419 | |
---|
420 | if ( pid == 0 ) |
---|
421 | { |
---|
422 | date = proctime(); |
---|
423 | PRINTF("\n*** Starting Instrumentation at cycle %d\n\n", date); |
---|
424 | |
---|
425 | int cc, pp; |
---|
426 | unsigned int min_load_start = 1000000000; |
---|
427 | unsigned int max_load_start = 0; |
---|
428 | unsigned int min_load_ended = 1000000000; |
---|
429 | unsigned int max_load_ended = 0; |
---|
430 | |
---|
431 | unsigned int min_hori_start = 1000000000; |
---|
432 | unsigned int max_hori_start = 0; |
---|
433 | unsigned int min_hori_ended = 1000000000; |
---|
434 | unsigned int max_hori_ended = 0; |
---|
435 | |
---|
436 | unsigned int min_vert_start = 1000000000; |
---|
437 | unsigned int max_vert_start = 0; |
---|
438 | unsigned int min_vert_ended = 1000000000; |
---|
439 | unsigned int max_vert_ended = 0; |
---|
440 | |
---|
441 | unsigned int min_disp_start = 1000000000; |
---|
442 | unsigned int max_disp_start = 0; |
---|
443 | unsigned int min_disp_ended = 1000000000; |
---|
444 | unsigned int max_disp_ended = 0; |
---|
445 | |
---|
446 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
447 | { |
---|
448 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
449 | { |
---|
450 | if ( LOAD_START[cc][pp] < min_load_start ) min_load_start = LOAD_START[cc][pp]; |
---|
451 | if ( LOAD_START[cc][pp] > max_load_start ) max_load_start = LOAD_START[cc][pp]; |
---|
452 | if ( LOAD_ENDED[cc][pp] < min_load_ended ) min_load_ended = LOAD_ENDED[cc][pp]; |
---|
453 | if ( LOAD_ENDED[cc][pp] > max_load_ended ) max_load_ended = LOAD_ENDED[cc][pp]; |
---|
454 | |
---|
455 | if ( HORI_START[cc][pp] < min_hori_start ) min_hori_start = HORI_START[cc][pp]; |
---|
456 | if ( HORI_START[cc][pp] > max_hori_start ) max_hori_start = HORI_START[cc][pp]; |
---|
457 | if ( HORI_ENDED[cc][pp] < min_hori_ended ) min_hori_ended = HORI_ENDED[cc][pp]; |
---|
458 | if ( HORI_ENDED[cc][pp] > max_hori_ended ) max_hori_ended = HORI_ENDED[cc][pp]; |
---|
459 | |
---|
460 | if ( VERT_START[cc][pp] < min_vert_start ) min_vert_start = VERT_START[cc][pp]; |
---|
461 | if ( VERT_START[cc][pp] > max_vert_start ) max_vert_start = VERT_START[cc][pp]; |
---|
462 | if ( VERT_ENDED[cc][pp] < min_vert_ended ) min_vert_ended = VERT_ENDED[cc][pp]; |
---|
463 | if ( VERT_ENDED[cc][pp] > max_vert_ended ) max_vert_ended = VERT_ENDED[cc][pp]; |
---|
464 | |
---|
465 | if ( DISP_START[cc][pp] < min_disp_start ) min_disp_start = DISP_START[cc][pp]; |
---|
466 | if ( DISP_START[cc][pp] > max_disp_start ) max_disp_start = DISP_START[cc][pp]; |
---|
467 | if ( DISP_ENDED[cc][pp] < min_disp_ended ) min_disp_ended = DISP_ENDED[cc][pp]; |
---|
468 | if ( DISP_ENDED[cc][pp] > max_disp_ended ) max_disp_ended = DISP_ENDED[cc][pp]; |
---|
469 | } |
---|
470 | } |
---|
471 | PRINTF(" - LOAD_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
472 | min_load_start, max_load_start, (min_load_start+max_load_start)/2, max_load_start-min_load_start); |
---|
473 | PRINTF(" - LOAD_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
474 | min_load_ended, max_load_ended, (min_load_ended+max_load_ended)/2, max_load_ended-min_load_ended); |
---|
475 | |
---|
476 | PRINTF(" - HORI_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
477 | min_hori_start, max_hori_start, (min_hori_start+max_hori_start)/2, max_hori_start-min_hori_start); |
---|
478 | PRINTF(" - HORI_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
479 | min_hori_ended, max_hori_ended, (min_hori_ended+max_hori_ended)/2, max_hori_ended-min_hori_ended); |
---|
480 | |
---|
481 | PRINTF(" - VERT_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
482 | min_vert_start, max_vert_start, (min_vert_start+max_vert_start)/2, max_vert_start-min_vert_start); |
---|
483 | PRINTF(" - VERT_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
484 | min_vert_ended, max_vert_ended, (min_vert_ended+max_vert_ended)/2, max_vert_ended-min_vert_ended); |
---|
485 | |
---|
486 | PRINTF(" - DISP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
487 | min_disp_start, max_disp_start, (min_disp_start+max_disp_start)/2, max_disp_start-min_disp_start); |
---|
488 | PRINTF(" - DISP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
489 | min_disp_ended, max_disp_ended, (min_disp_ended+max_disp_ended)/2, max_disp_ended-min_disp_ended); |
---|
490 | |
---|
491 | PRINTF(" - BARRIER LOAD/HORI = %d\n", min_hori_start - max_load_ended); |
---|
492 | PRINTF(" - BARRIER HORI/VERT = %d\n", min_vert_start - max_hori_ended); |
---|
493 | PRINTF(" - BARRIER VERT/DISP = %d\n", min_disp_start - max_vert_ended); |
---|
494 | |
---|
495 | PRINTF(" - LOAD = %d\n", max_load_ended); |
---|
496 | PRINTF(" - FILTER = %d\n", max_vert_ended - max_load_ended); |
---|
497 | PRINTF(" - DISPLAY = %d\n", max_disp_ended - max_vert_ended); |
---|
498 | |
---|
499 | PRINTF("\nBEGIN LOAD_START\n"); |
---|
500 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
501 | { |
---|
502 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
503 | { |
---|
504 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, LOAD_START[cc][pp]); |
---|
505 | } |
---|
506 | } |
---|
507 | PRINTF("END\n"); |
---|
508 | PRINTF("\nBEGIN LOAD_ENDED\n"); |
---|
509 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
510 | { |
---|
511 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
512 | { |
---|
513 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, LOAD_ENDED[cc][pp]); |
---|
514 | } |
---|
515 | } |
---|
516 | PRINTF("END\n"); |
---|
517 | PRINTF("\nBEGIN HORI_START\n"); |
---|
518 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
519 | { |
---|
520 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
521 | { |
---|
522 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, HORI_START[cc][pp]); |
---|
523 | } |
---|
524 | } |
---|
525 | PRINTF("END\n"); |
---|
526 | PRINTF("\nBEGIN HORI_ENDED\n"); |
---|
527 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
528 | { |
---|
529 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
530 | { |
---|
531 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, HORI_ENDED[cc][pp]); |
---|
532 | } |
---|
533 | } |
---|
534 | PRINTF("END\n"); |
---|
535 | PRINTF("\nBEGIN VERT_START\n"); |
---|
536 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
537 | { |
---|
538 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
539 | { |
---|
540 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, VERT_START[cc][pp]); |
---|
541 | } |
---|
542 | } |
---|
543 | PRINTF("END\n"); |
---|
544 | PRINTF("\nBEGIN VERT_ENDED\n"); |
---|
545 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
546 | { |
---|
547 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
548 | { |
---|
549 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, VERT_ENDED[cc][pp]); |
---|
550 | } |
---|
551 | } |
---|
552 | PRINTF("END\n"); |
---|
553 | PRINTF("\nBEGIN DISP_START\n"); |
---|
554 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
555 | { |
---|
556 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
557 | { |
---|
558 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, DISP_START[cc][pp]); |
---|
559 | } |
---|
560 | } |
---|
561 | PRINTF("END\n"); |
---|
562 | PRINTF("\nBEGIN DISP_ENDED\n"); |
---|
563 | for ( cc=0 ; cc<nclusters ; cc++ ) |
---|
564 | { |
---|
565 | for ( pp=0 ; pp<nprocs ; pp++ ) |
---|
566 | { |
---|
567 | PRINTF("cluster= %d proc= %d date= %d\n", cc, pp, DISP_ENDED[cc][pp]); |
---|
568 | } |
---|
569 | } |
---|
570 | PRINTF("END\n"); |
---|
571 | } |
---|
572 | |
---|
573 | while(1); |
---|
574 | |
---|
575 | } // end main() |
---|
576 | |
---|