1 | ////////////////////////////////////////////////////////////////////////////////////////// |
---|
2 | // File : transpose.c |
---|
3 | // Date : september 2019 |
---|
4 | // author : Alain Greiner |
---|
5 | ////////////////////////////////////////////////////////////////////////////////////////// |
---|
6 | // This multi-threaded aplication read a raw image (one byte per pixel) |
---|
7 | // stored on disk, transpose it, display the result on the frame buffer, |
---|
8 | // and store the transposed image on disk. |
---|
9 | // It can run on a multi-cores, multi-clusters architecture, with one thread |
---|
10 | // |
---|
11 | // per core, and uses the POSIX threads API. |
---|
12 | // It uses the mmap() syscall to directly access the input and output files |
---|
13 | // and the fbf_write() syscall to display the images. |
---|
14 | // |
---|
15 | // The main() function can be launched on any core[cxy,l]. |
---|
16 | // It makes the initialisations, launch (N-1) threads to run the execute() function |
---|
17 | // on the (N-1) other cores, calls himself the execute() function, and finally calls |
---|
18 | // the instrument() function to display instrumentation results when the parallel |
---|
19 | // execution is completed. The placement of threads on the cores can be done |
---|
20 | // automatically by the operating system, or can be done explicitely by the main thread |
---|
21 | // (when the EXPLICIT_PLACEMENT global parameter is set). |
---|
22 | // |
---|
23 | // The buf_in[x,y] and buf_out[put buffers containing the direct ans transposed images |
---|
24 | // are distributed in clusters: In each cluster[cxy], the thread running on core[cxy,0] |
---|
25 | // map the buf_in[cxy] and // buf_out[cxy] buffers containing a subset of lines. |
---|
26 | // Then, all threads in cluster[xy] read pixels from the local buf_in[cxy] buffer, and |
---|
27 | // write the pixels to all remote buf_out[cxy] buffers. Finally, each thread display |
---|
28 | // a part of the transposed image to the frame buffer. |
---|
29 | // |
---|
30 | // - The image must fit the frame buffer size, that must be power of 2. |
---|
31 | // - The number of clusters must be a power of 2 no larger than 256. |
---|
32 | // - The number of cores per cluster must be a power of 2 no larger than 4. |
---|
33 | // - The number of clusters cannot be larger than (IMAGE_SIZE * IMAGE_SIZE) / 4096, |
---|
34 | // because the size of buf_in[x,y] and buf_out[x,y] must be multiple of 4096. |
---|
35 | // |
---|
36 | ////////////////////////////////////////////////////////////////////////////////////////// |
---|
37 | |
---|
38 | #include <sys/mman.h> |
---|
39 | #include <stdio.h> |
---|
40 | #include <stdlib.h> |
---|
41 | #include <unistd.h> |
---|
42 | #include <pthread.h> |
---|
43 | #include <string.h> |
---|
44 | #include <almosmkh.h> |
---|
45 | #include <fcntl.h> |
---|
46 | #include <hal_macros.h> |
---|
47 | |
---|
48 | #define X_MAX 16 // max number of clusters in row |
---|
49 | #define Y_MAX 16 // max number of clusters in column |
---|
50 | #define CORES_MAX 4 // max number of cores per cluster |
---|
51 | #define CLUSTERS_MAX (X_MAX * Y_MAX) // max number of clusters |
---|
52 | |
---|
53 | #define IMAGE_SIZE 256 // image size |
---|
54 | #define IMAGE_TYPE 420 // pixel encoding type |
---|
55 | #define INPUT_FILE_PATH "/misc/lena_256.raw" // input file pathname |
---|
56 | #define OUTPUT_FILE_PATH "/home/trsp_256.raw" // output file pathname |
---|
57 | |
---|
58 | #define USE_DQT_BARRIER 1 // quad-tree barrier if non zero |
---|
59 | #define EXPLICIT_PLACEMENT 1 // explicit thread placement |
---|
60 | #define VERBOSE 1 // print comments on TTY |
---|
61 | |
---|
62 | |
---|
63 | /////////////////////////////////////////////////////// |
---|
64 | // global variables |
---|
65 | /////////////////////////////////////////////////////// |
---|
66 | |
---|
67 | // instrumentation counters for each processor in each cluster |
---|
68 | unsigned int MMAP_START[CLUSTERS_MAX][CORES_MAX] = {{ 0 }}; |
---|
69 | unsigned int MMAP_END [CLUSTERS_MAX][CORES_MAX] = {{ 0 }}; |
---|
70 | unsigned int TRSP_START[CLUSTERS_MAX][CORES_MAX] = {{ 0 }}; |
---|
71 | unsigned int TRSP_END [CLUSTERS_MAX][CORES_MAX] = {{ 0 }}; |
---|
72 | unsigned int DISP_START[CLUSTERS_MAX][CORES_MAX] = {{ 0 }}; |
---|
73 | unsigned int DISP_END [CLUSTERS_MAX][CORES_MAX] = {{ 0 }}; |
---|
74 | |
---|
75 | // arrays of pointers on distributed buffers |
---|
76 | // one input buffer & one output buffer per cluster |
---|
77 | unsigned char * buf_in [CLUSTERS_MAX]; |
---|
78 | unsigned char * buf_out[CLUSTERS_MAX]; |
---|
79 | |
---|
80 | // synchronisation barrier (all threads) |
---|
81 | pthread_barrier_t barrier; |
---|
82 | |
---|
83 | // platform parameters |
---|
84 | unsigned int x_size; // number of clusters in a row |
---|
85 | unsigned int y_size; // number of clusters in a column |
---|
86 | unsigned int ncores; // number of processors per cluster |
---|
87 | |
---|
88 | // cluster identifier & local index of core running the main thread |
---|
89 | unsigned int cxy_main; |
---|
90 | unsigned int lid_main; |
---|
91 | |
---|
92 | // input & output file descriptors |
---|
93 | int fd_in; |
---|
94 | int fd_out; |
---|
95 | |
---|
96 | #if EXPLICIT_PLACEMENT |
---|
97 | |
---|
98 | // thread index allocated by the kernel |
---|
99 | pthread_t trdid[CLUSTERS_MAX][CORES_MAX]; |
---|
100 | |
---|
101 | // user defined continuous thread index |
---|
102 | unsigned int tid[CLUSTERS_MAX][CORES_MAX]; |
---|
103 | |
---|
104 | // thread attributes only used if explicit placement |
---|
105 | pthread_attr_t attr[CLUSTERS_MAX][CORES_MAX]; |
---|
106 | |
---|
107 | #else |
---|
108 | |
---|
109 | // thread index allocated by the kernel |
---|
110 | pthread_t trdid[CLUSTERS_MAX * CORES_MAX]; |
---|
111 | |
---|
112 | // user defined continuous thread index |
---|
113 | unsigned int tid[CLUSTERS_MAX * CORES_MAX]; |
---|
114 | |
---|
115 | #endif |
---|
116 | |
---|
117 | //return values at thread exit |
---|
118 | unsigned int THREAD_EXIT_SUCCESS = 0; |
---|
119 | unsigned int THREAD_EXIT_FAILURE = 1; |
---|
120 | |
---|
121 | //////////////////////////////////////////////////////////////// |
---|
122 | // functions declaration |
---|
123 | //////////////////////////////////////////////////////////////// |
---|
124 | |
---|
125 | void execute( unsigned int * ptid ); |
---|
126 | |
---|
127 | void instrument( void ); |
---|
128 | |
---|
129 | /////////// |
---|
130 | void main() |
---|
131 | { |
---|
132 | unsigned long long date; |
---|
133 | |
---|
134 | int error; |
---|
135 | |
---|
136 | printf("\n bloup 0\n"); |
---|
137 | |
---|
138 | // get identifiers for core executing main |
---|
139 | get_core_id( &cxy_main , &lid_main ); |
---|
140 | |
---|
141 | printf("\n bloup 1\n"); |
---|
142 | |
---|
143 | // get & check plat-form parameters |
---|
144 | get_config( &x_size , &y_size , &ncores ); |
---|
145 | |
---|
146 | printf("\n bloup 2\n"); |
---|
147 | |
---|
148 | if((ncores != 1) && (ncores != 2) && (ncores == 4)) |
---|
149 | { |
---|
150 | printf("\n[transpose error] number of cores per cluster must be 1/2/4\n"); |
---|
151 | exit( 0 ); |
---|
152 | } |
---|
153 | |
---|
154 | if( (x_size != 1) && (x_size != 2) && (x_size != 4) && |
---|
155 | (x_size != 8) && (x_size != 16) ) |
---|
156 | { |
---|
157 | printf("\n[transpose error] x_size must be 1/2/4/8/16\n"); |
---|
158 | exit( 0 ); |
---|
159 | } |
---|
160 | |
---|
161 | if( (y_size != 1) && (y_size != 2) && (y_size != 4) && |
---|
162 | (y_size != 8) && (y_size != 16) ) |
---|
163 | { |
---|
164 | printf("\n[transpose error] y_size must be 1/2/4/8/16\n"); |
---|
165 | exit( 0 ); |
---|
166 | } |
---|
167 | |
---|
168 | printf("\n bloup 3\n"); |
---|
169 | |
---|
170 | // compute number of threads |
---|
171 | unsigned int nclusters = x_size * y_size; |
---|
172 | unsigned int nthreads = nclusters * ncores; |
---|
173 | |
---|
174 | printf("\n bloup 4\n"); |
---|
175 | |
---|
176 | // get FBF ownership and FBF size |
---|
177 | unsigned int fbf_width; |
---|
178 | unsigned int fbf_height; |
---|
179 | unsigned int fbf_type; |
---|
180 | fbf_get_config( &fbf_width , &fbf_height , &fbf_type ); |
---|
181 | |
---|
182 | printf("\n bloup 5\n"); |
---|
183 | |
---|
184 | if( (fbf_width != IMAGE_SIZE) || (fbf_height != IMAGE_SIZE) || (fbf_type != IMAGE_TYPE) ) |
---|
185 | { |
---|
186 | printf("\n[transpose error] image does not fit FBF size or type\n"); |
---|
187 | exit( 0 ); |
---|
188 | } |
---|
189 | |
---|
190 | get_cycle( &date ); |
---|
191 | printf("\n[transpose] starts at cycle %d on %d cores / FBF = %d * %d pixels\n", |
---|
192 | (unsigned int)date , nthreads , fbf_width , fbf_height ); |
---|
193 | |
---|
194 | // open input file |
---|
195 | fd_in = open( INPUT_FILE_PATH , O_RDONLY , 0 ); // read-only |
---|
196 | if ( fd_in < 0 ) |
---|
197 | { |
---|
198 | printf("\n[transpose error] main cannot open file %s\n", INPUT_FILE_PATH ); |
---|
199 | exit( 0 ); |
---|
200 | } |
---|
201 | |
---|
202 | #if VERBOSE |
---|
203 | printf("\n[transpose] main open file %s / fd = %d\n", INPUT_FILE_PATH , fd_in ); |
---|
204 | #endif |
---|
205 | |
---|
206 | // open output file |
---|
207 | fd_out = open( OUTPUT_FILE_PATH , O_CREAT , 0 ); // create if required |
---|
208 | if ( fd_out < 0 ) |
---|
209 | { |
---|
210 | printf("\n[transpose error] main cannot open file %s\n", OUTPUT_FILE_PATH ); |
---|
211 | exit( 0 ); |
---|
212 | } |
---|
213 | |
---|
214 | #if VERBOSE |
---|
215 | printf("\n[transpose] main open file %s / fd = %d\n", OUTPUT_FILE_PATH , fd_out ); |
---|
216 | #endif |
---|
217 | |
---|
218 | // initialise barrier |
---|
219 | if( USE_DQT_BARRIER ) |
---|
220 | { |
---|
221 | pthread_barrierattr_t attr; |
---|
222 | attr.x_size = x_size; |
---|
223 | attr.y_size = y_size; |
---|
224 | attr.nthreads = ncores; |
---|
225 | error = pthread_barrier_init( &barrier, &attr , nthreads ); |
---|
226 | } |
---|
227 | else |
---|
228 | { |
---|
229 | error = pthread_barrier_init( &barrier, NULL , nthreads ); |
---|
230 | } |
---|
231 | |
---|
232 | if( error ) |
---|
233 | { |
---|
234 | printf("\n[transpose error] main cannot initialize barrier\n" ); |
---|
235 | exit( 0 ); |
---|
236 | } |
---|
237 | |
---|
238 | get_cycle( &date ); |
---|
239 | printf("\n[transpose] main on core[%x,%d] completes initialisation at cycle %d\n" |
---|
240 | "- CLUSTERS = %d\n" |
---|
241 | "- PROCS = %d\n" |
---|
242 | "- THREADS = %d\n", |
---|
243 | cxy_main, lid_main, (unsigned int)date, nclusters, ncores, nthreads ); |
---|
244 | |
---|
245 | ////////////////////// |
---|
246 | #if EXPLICIT_PLACEMENT |
---|
247 | |
---|
248 | // main thread launch other threads |
---|
249 | unsigned int x; |
---|
250 | unsigned int y; |
---|
251 | unsigned int l; |
---|
252 | unsigned int cxy; |
---|
253 | for( x = 0 ; x < x_size ; x++ ) |
---|
254 | { |
---|
255 | for( y = 0 ; y < y_size ; y++ ) |
---|
256 | { |
---|
257 | cxy = HAL_CXY_FROM_XY( x , y ); |
---|
258 | for( l = 0 ; l < ncores ; l++ ) |
---|
259 | { |
---|
260 | // no other thread on the core running the main |
---|
261 | if( (cxy != cxy_main) || (l != lid_main) ) |
---|
262 | { |
---|
263 | // define thread attributes |
---|
264 | attr[cxy][l].attributes = PT_ATTR_CLUSTER_DEFINED | PT_ATTR_CORE_DEFINED; |
---|
265 | attr[cxy][l].cxy = cxy; |
---|
266 | attr[cxy][l].lid = l; |
---|
267 | |
---|
268 | tid[cxy][l] = (((x * y_size) + y) * ncores) + l; |
---|
269 | |
---|
270 | // create thread on core[cxy,l] |
---|
271 | if (pthread_create( &trdid[cxy][l], |
---|
272 | &attr[cxy][l], |
---|
273 | &execute, |
---|
274 | &tid[cxy][l] ) ) |
---|
275 | { |
---|
276 | printf("\n[convol error] created thread %x on core[%x][%d]\n", |
---|
277 | trdid[cxy][l] , cxy , l ); |
---|
278 | exit( 0 ); |
---|
279 | } |
---|
280 | #if VERBOSE |
---|
281 | printf("\n[transpose] main created thread[%x,%d]\n", cxy, l ); |
---|
282 | #endif |
---|
283 | } |
---|
284 | } |
---|
285 | } |
---|
286 | } |
---|
287 | |
---|
288 | // main thread calls itself the execute() function |
---|
289 | execute( &tid[cxy_main][lid_main] ); |
---|
290 | |
---|
291 | // main thread wait other threads completion |
---|
292 | for( x = 0 ; x < x_size ; x++ ) |
---|
293 | { |
---|
294 | for( y = 0 ; y < y_size ; y++ ) |
---|
295 | { |
---|
296 | cxy = HAL_CXY_FROM_XY( x , y ); |
---|
297 | for( l = 0 ; l < ncores ; l++ ) |
---|
298 | { |
---|
299 | // no other thread on the core running the main |
---|
300 | if( (cxy != cxy_main) || (l != lid_main) ) |
---|
301 | { |
---|
302 | unsigned int * status; |
---|
303 | |
---|
304 | // wait thread[cxy][l] |
---|
305 | if( pthread_join( trdid[cxy][l] , (void*)(&status) ) ) |
---|
306 | { |
---|
307 | printf("\n[transpose error] main cannot join thread[%x,%d]\n", cxy, l ); |
---|
308 | exit( 0 ); |
---|
309 | } |
---|
310 | |
---|
311 | // check status |
---|
312 | if( *status != THREAD_EXIT_SUCCESS ) |
---|
313 | { |
---|
314 | printf("\n[transpose error] thread[%x,%d] returned failure\n", cxy, l ); |
---|
315 | exit( 0 ); |
---|
316 | } |
---|
317 | #if VERBOSE |
---|
318 | printf("\n[transpose] main joined thread[%x,%d]\n", cxy, l ); |
---|
319 | #endif |
---|
320 | } |
---|
321 | } |
---|
322 | } |
---|
323 | } |
---|
324 | |
---|
325 | /////////////////////////////// |
---|
326 | #else // no explicit placement |
---|
327 | |
---|
328 | // main thread launch other threads |
---|
329 | unsigned int n; |
---|
330 | for ( n = 1 ; n < nthreads ; n++ ) |
---|
331 | { |
---|
332 | tid[n] = n; |
---|
333 | if ( pthread_create( &trdid[n], |
---|
334 | NULL, // no attribute |
---|
335 | &execute, |
---|
336 | &tid[n] ) ) |
---|
337 | { |
---|
338 | printf("\n[transpose error] cannot create thread %d\n", n ); |
---|
339 | exit( 0 ); |
---|
340 | } |
---|
341 | |
---|
342 | #if VERBOSE |
---|
343 | printf("\n[transpose] main created thread %d\n", tid[n] ); |
---|
344 | #endif |
---|
345 | |
---|
346 | } |
---|
347 | |
---|
348 | // main thread calls itself the execute() function |
---|
349 | execute( &tid[0] ); |
---|
350 | |
---|
351 | // main thread wait other threads completion |
---|
352 | for ( n = 1 ; n < nthreads ; n++ ) |
---|
353 | { |
---|
354 | unsigned int * status; |
---|
355 | |
---|
356 | // main wait thread[n] status |
---|
357 | if ( pthread_join( trdid[n], (void*)(&status)) ) |
---|
358 | { |
---|
359 | printf("\n[transpose error] main cannot join thread %d\n", n ); |
---|
360 | exit( 0 ); |
---|
361 | } |
---|
362 | |
---|
363 | // check status |
---|
364 | if( *status != THREAD_EXIT_SUCCESS ) |
---|
365 | { |
---|
366 | printf("\n[transpose error] thread %x returned failure\n", n ); |
---|
367 | exit( 0 ); |
---|
368 | } |
---|
369 | |
---|
370 | #if VERBOSE |
---|
371 | printf("\n[transpose] main successfully joined thread %x\n", tid[n] ); |
---|
372 | #endif |
---|
373 | |
---|
374 | } |
---|
375 | |
---|
376 | #endif |
---|
377 | |
---|
378 | // instrumentation |
---|
379 | instrument(); |
---|
380 | |
---|
381 | // close input and output files |
---|
382 | close( fd_in ); |
---|
383 | close( fd_out ); |
---|
384 | |
---|
385 | // suicide |
---|
386 | exit( 0 ); |
---|
387 | |
---|
388 | } // end main() |
---|
389 | |
---|
390 | |
---|
391 | |
---|
392 | /////////////////////////////////// |
---|
393 | void execute( unsigned int * ptid ) |
---|
394 | { |
---|
395 | unsigned long long date; |
---|
396 | |
---|
397 | unsigned int l; // line index for loops |
---|
398 | unsigned int p; // pixel index for loops |
---|
399 | |
---|
400 | // get thread continuous index |
---|
401 | unsigned int my_tid = *ptid; |
---|
402 | |
---|
403 | // build total number of pixels per image |
---|
404 | unsigned int npixels = IMAGE_SIZE * IMAGE_SIZE; |
---|
405 | |
---|
406 | // nuild total number of threads and clusters |
---|
407 | unsigned int nthreads = x_size * y_size * ncores; |
---|
408 | unsigned int nclusters = x_size * y_size; |
---|
409 | |
---|
410 | // get cluster continuous index and core index from tid |
---|
411 | // we use (tid == cid * ncores + lid) |
---|
412 | unsigned int cid = my_tid / ncores; // continuous index |
---|
413 | unsigned int lid = my_tid % ncores; // core local index |
---|
414 | |
---|
415 | // get cluster identifier from cid |
---|
416 | // we use (cid == x * y_size + y) |
---|
417 | unsigned int x = cid / y_size; // X cluster coordinate |
---|
418 | unsigned int y = cid % y_size; // Y cluster coordinate |
---|
419 | unsigned int cxy = HAL_CXY_FROM_XY(x,y); |
---|
420 | |
---|
421 | #if VERBOSE |
---|
422 | printf("\n[transpose] thread[%d] start on core[%x,%d]\n", my_tid , cxy , lid ); |
---|
423 | #endif |
---|
424 | |
---|
425 | // In each cluster cxy, thread[cxy,0] map input file |
---|
426 | // to buf_in[cxy] and map output file to buf_in[cxy] |
---|
427 | |
---|
428 | get_cycle( &date ); |
---|
429 | MMAP_START[cxy][lid] = (unsigned int)date; |
---|
430 | |
---|
431 | if ( lid == 0 ) |
---|
432 | { |
---|
433 | unsigned int length = npixels / nclusters; |
---|
434 | unsigned int offset = length * cid; |
---|
435 | |
---|
436 | // map buf_in |
---|
437 | buf_in[cid] = mmap( NULL, |
---|
438 | length, |
---|
439 | PROT_READ, |
---|
440 | MAP_SHARED, |
---|
441 | fd_in, |
---|
442 | offset ); |
---|
443 | |
---|
444 | if ( buf_in[cid] == NULL ) |
---|
445 | { |
---|
446 | printf("\n[transpose error] thread[%x,%d] cannot map input file\n", cxy, lid); |
---|
447 | pthread_exit( &THREAD_EXIT_FAILURE ); |
---|
448 | } |
---|
449 | |
---|
450 | #if VERBOSE |
---|
451 | printf("\n[transpose] thread[%x,%d] map input file / length %x / offset %x / buf_in %x\n", |
---|
452 | cxy, lid, length, offset, buf_in[cid] ); |
---|
453 | #endif |
---|
454 | |
---|
455 | // map buf_out |
---|
456 | buf_out[cid] = mmap( NULL, |
---|
457 | length, |
---|
458 | PROT_WRITE, |
---|
459 | MAP_SHARED, |
---|
460 | fd_out, |
---|
461 | offset ); |
---|
462 | |
---|
463 | if ( buf_out[cid] == NULL ) |
---|
464 | { |
---|
465 | printf("\n[transpose error] thread[%x,%d] cannot map output file\n", cxy, lid); |
---|
466 | pthread_exit( &THREAD_EXIT_FAILURE ); |
---|
467 | } |
---|
468 | |
---|
469 | #if VERBOSE |
---|
470 | printf("\n[transpose] thread[%x,%d] map output file / length %x / offset %x / buf_out %x\n", |
---|
471 | cxy, lid, length, offset, buf_out[cid] ); |
---|
472 | #endif |
---|
473 | |
---|
474 | } |
---|
475 | |
---|
476 | get_cycle( &date ); |
---|
477 | MMAP_END[cxy][lid] = (unsigned int)date; |
---|
478 | |
---|
479 | ///////////////////////////////// |
---|
480 | pthread_barrier_wait( &barrier ); |
---|
481 | |
---|
482 | // parallel transpose from buf_in to buf_out |
---|
483 | // each thread makes the transposition for nlt lines (nlt = IMAGE_SIZE/nthreads) |
---|
484 | // from line [tid*nlt] to line [(tid + 1)*nlt - 1] |
---|
485 | // (p,l) are the absolute pixel coordinates in the source image |
---|
486 | |
---|
487 | get_cycle( &date ); |
---|
488 | TRSP_START[cxy][lid] = (unsigned int)date; |
---|
489 | |
---|
490 | unsigned int nlt = IMAGE_SIZE / nthreads; // number of lines per thread |
---|
491 | unsigned int nlc = IMAGE_SIZE / nclusters; // number of lines per cluster |
---|
492 | |
---|
493 | unsigned int src_cluster; |
---|
494 | unsigned int src_index; |
---|
495 | unsigned int dst_cluster; |
---|
496 | unsigned int dst_index; |
---|
497 | |
---|
498 | unsigned char byte; |
---|
499 | |
---|
500 | unsigned int first = my_tid * nlt; // first line index for a given thread |
---|
501 | unsigned int last = first + nlt; // last line index for a given thread |
---|
502 | |
---|
503 | for ( l = first ; l < last ; l++ ) |
---|
504 | { |
---|
505 | // in each iteration we transfer one byte |
---|
506 | for ( p = 0 ; p < IMAGE_SIZE ; p++ ) |
---|
507 | { |
---|
508 | // read one byte from local buf_in |
---|
509 | src_cluster = l / nlc; |
---|
510 | src_index = (l % nlc) * IMAGE_SIZE + p; |
---|
511 | byte = buf_in[src_cluster][src_index]; |
---|
512 | |
---|
513 | // write one byte to remote buf_out |
---|
514 | dst_cluster = p / nlc; |
---|
515 | dst_index = (p % nlc) * IMAGE_SIZE + l; |
---|
516 | |
---|
517 | buf_out[dst_cluster][dst_index] = byte; |
---|
518 | } |
---|
519 | } |
---|
520 | |
---|
521 | #if VERBOSE |
---|
522 | printf("\n[transpose] thread[%x,%d] completes transposed\n", cxy, lid ); |
---|
523 | #endif |
---|
524 | |
---|
525 | get_cycle( &date ); |
---|
526 | TRSP_END[cxy][lid] = (unsigned int)date; |
---|
527 | |
---|
528 | ///////////////////////////////// |
---|
529 | pthread_barrier_wait( &barrier ); |
---|
530 | |
---|
531 | // parallel display from local buf_out to frame buffer |
---|
532 | // all threads contribute to display |
---|
533 | |
---|
534 | get_cycle( &date ); |
---|
535 | DISP_START[cxy][lid] = (unsigned int)date; |
---|
536 | |
---|
537 | unsigned int npt = npixels / nthreads; // number of pixels per thread |
---|
538 | |
---|
539 | if( fbf_write( &buf_out[cid][lid * npt], |
---|
540 | npt, |
---|
541 | npt * my_tid ) ) |
---|
542 | { |
---|
543 | printf("\n[transpose error] thread[%x,%d] cannot access FBF\n", cxy, lid ); |
---|
544 | pthread_exit( &THREAD_EXIT_FAILURE ); |
---|
545 | } |
---|
546 | |
---|
547 | #if VERBOSE |
---|
548 | printf("\n[transpose] thread[%x,%d] completes display\n", cxy, lid ); |
---|
549 | #endif |
---|
550 | |
---|
551 | get_cycle( &date ); |
---|
552 | DISP_END[cxy][lid] = (unsigned int)date; |
---|
553 | |
---|
554 | ///////////////////////////////// |
---|
555 | pthread_barrier_wait( &barrier ); |
---|
556 | |
---|
557 | // all threads, but thread[0,0,0], suicide |
---|
558 | if ( (cxy != cxy_main) || (lid != lid_main) ) |
---|
559 | { |
---|
560 | pthread_exit( &THREAD_EXIT_SUCCESS ); |
---|
561 | } |
---|
562 | |
---|
563 | } // end execute() |
---|
564 | |
---|
565 | |
---|
566 | |
---|
567 | /////////////////////// |
---|
568 | void instrument( void ) |
---|
569 | { |
---|
570 | unsigned int x, y, l; |
---|
571 | |
---|
572 | unsigned int min_load_start = 0xFFFFFFFF; |
---|
573 | unsigned int max_load_start = 0; |
---|
574 | unsigned int min_load_ended = 0xFFFFFFFF; |
---|
575 | unsigned int max_load_ended = 0; |
---|
576 | unsigned int min_trsp_start = 0xFFFFFFFF; |
---|
577 | unsigned int max_trsp_start = 0; |
---|
578 | unsigned int min_trsp_ended = 0xFFFFFFFF; |
---|
579 | unsigned int max_trsp_ended = 0; |
---|
580 | unsigned int min_disp_start = 0xFFFFFFFF; |
---|
581 | unsigned int max_disp_start = 0; |
---|
582 | unsigned int min_disp_ended = 0xFFFFFFFF; |
---|
583 | unsigned int max_disp_ended = 0; |
---|
584 | |
---|
585 | char string[64]; |
---|
586 | |
---|
587 | snprintf( string , 64 , "/home/transpose_%d_%d_%d" , x_size , y_size , ncores ); |
---|
588 | |
---|
589 | // open instrumentation file |
---|
590 | FILE * f = fopen( string , NULL ); |
---|
591 | if ( f == NULL ) |
---|
592 | { |
---|
593 | printf("\n[transpose error] cannot open instrumentation file %s\n", string ); |
---|
594 | exit( 0 ); |
---|
595 | } |
---|
596 | |
---|
597 | for (x = 0; x < x_size; x++) |
---|
598 | { |
---|
599 | for (y = 0; y < y_size; y++) |
---|
600 | { |
---|
601 | unsigned int cxy = HAL_CXY_FROM_XY( x , y ); |
---|
602 | |
---|
603 | for ( l = 0 ; l < ncores ; l++ ) |
---|
604 | { |
---|
605 | if (MMAP_START[cxy][l] < min_load_start) min_load_start = MMAP_START[cxy][l]; |
---|
606 | if (MMAP_START[cxy][l] > max_load_start) max_load_start = MMAP_START[cxy][l]; |
---|
607 | if (MMAP_END[cxy][l] < min_load_ended) min_load_ended = MMAP_END[cxy][l]; |
---|
608 | if (MMAP_END[cxy][l] > max_load_ended) max_load_ended = MMAP_END[cxy][l]; |
---|
609 | if (TRSP_START[cxy][l] < min_trsp_start) min_trsp_start = TRSP_START[cxy][l]; |
---|
610 | if (TRSP_START[cxy][l] > max_trsp_start) max_trsp_start = TRSP_START[cxy][l]; |
---|
611 | if (TRSP_END[cxy][l] < min_trsp_ended) min_trsp_ended = TRSP_END[cxy][l]; |
---|
612 | if (TRSP_END[cxy][l] > max_trsp_ended) max_trsp_ended = TRSP_END[cxy][l]; |
---|
613 | if (DISP_START[cxy][l] < min_disp_start) min_disp_start = DISP_START[cxy][l]; |
---|
614 | if (DISP_START[cxy][l] > max_disp_start) max_disp_start = DISP_START[cxy][l]; |
---|
615 | if (DISP_END[cxy][l] < min_disp_ended) min_disp_ended = DISP_END[cxy][l]; |
---|
616 | if (DISP_END[cxy][l] > max_disp_ended) max_disp_ended = DISP_END[cxy][l]; |
---|
617 | } |
---|
618 | } |
---|
619 | } |
---|
620 | |
---|
621 | printf( "\n ------ %s ------\n" , string ); |
---|
622 | fprintf( f , "\n ------ %s ------\n" , string ); |
---|
623 | |
---|
624 | printf( " - MMAP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
625 | min_load_start, max_load_start, (min_load_start+max_load_start)/2, |
---|
626 | max_load_start-min_load_start ); |
---|
627 | |
---|
628 | fprintf( f , " - MMAP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
629 | min_load_start, max_load_start, (min_load_start+max_load_start)/2, |
---|
630 | max_load_start-min_load_start ); |
---|
631 | |
---|
632 | printf( " - MMAP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
633 | min_load_ended, max_load_ended, (min_load_ended+max_load_ended)/2, |
---|
634 | max_load_ended-min_load_ended ); |
---|
635 | |
---|
636 | fprintf( f , " - MMAP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
637 | min_load_ended, max_load_ended, (min_load_ended+max_load_ended)/2, |
---|
638 | max_load_ended-min_load_ended ); |
---|
639 | |
---|
640 | printf( " - TRSP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
641 | min_trsp_start, max_trsp_start, (min_trsp_start+max_trsp_start)/2, |
---|
642 | max_trsp_start-min_trsp_start ); |
---|
643 | |
---|
644 | fprintf( f , " - TRSP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
645 | min_trsp_start, max_trsp_start, (min_trsp_start+max_trsp_start)/2, |
---|
646 | max_trsp_start-min_trsp_start ); |
---|
647 | |
---|
648 | printf( " - TRSP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
649 | min_trsp_ended, max_trsp_ended, (min_trsp_ended+max_trsp_ended)/2, |
---|
650 | max_trsp_ended-min_trsp_ended ); |
---|
651 | |
---|
652 | fprintf( f , " - TRSP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
653 | min_trsp_ended, max_trsp_ended, (min_trsp_ended+max_trsp_ended)/2, |
---|
654 | max_trsp_ended-min_trsp_ended ); |
---|
655 | |
---|
656 | printf( " - DISP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
657 | min_disp_start, max_disp_start, (min_disp_start+max_disp_start)/2, |
---|
658 | max_disp_start-min_disp_start ); |
---|
659 | |
---|
660 | fprintf( f , " - DISP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
661 | min_disp_start, max_disp_start, (min_disp_start+max_disp_start)/2, |
---|
662 | max_disp_start-min_disp_start ); |
---|
663 | |
---|
664 | printf( " - DISP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
665 | min_disp_ended, max_disp_ended, (min_disp_ended+max_disp_ended)/2, |
---|
666 | max_disp_ended-min_disp_ended ); |
---|
667 | |
---|
668 | fprintf( f , " - DISP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
669 | min_disp_ended, max_disp_ended, (min_disp_ended+max_disp_ended)/2, |
---|
670 | max_disp_ended-min_disp_ended ); |
---|
671 | |
---|
672 | fclose( f ); |
---|
673 | |
---|
674 | } // end instrument() |
---|
675 | |
---|
676 | |
---|
677 | |
---|
678 | |
---|