1 | /////////////////////////////////////////////////////////////////////////////////////// |
---|
2 | // File : transpose.c |
---|
3 | // Date : september 2015 |
---|
4 | // author : Alain Greiner |
---|
5 | /////////////////////////////////////////////////////////////////////////////////////// |
---|
6 | // This multi-threaded aplication transposes a raw image (one pbyte per pixel). |
---|
7 | // It can run on a multi-processors, multi-clusters architecture, with one thread |
---|
8 | // per processor, and uses the POSIX threads API. |
---|
9 | // It does not use the CMA to display the result image. |
---|
10 | // |
---|
11 | // The main() function can be launched on any processor P[x,y,l]. |
---|
12 | // It makes the initialisations, launch (N-1) threads to run the execute() function |
---|
13 | // on the (N-1) other processors than P[x,y,l], call himself the execute() function, |
---|
14 | // and finally call the instrument() function to display instrumentation results |
---|
15 | // when the parallel execution is completed. |
---|
16 | // |
---|
17 | // The input and output buffers containing the image are distributed in clusters. |
---|
18 | // |
---|
19 | // The execute() function read a set of lines from an input file on disk, |
---|
20 | // to the local buffer buf_in[x][y], transpose it, write the result to a remote buffer |
---|
21 | // buf_out[x'][y'], display the content of the local buffer buf_out[x][y] to the |
---|
22 | // frame buffer, and store it on disk to another output file. |
---|
23 | // |
---|
24 | // - The image size must fit the frame buffer size. |
---|
25 | // - The block size in block device must be 512 bytes. |
---|
26 | // - The number of clusters must be a power of 2 no larger than 256. |
---|
27 | // - The number of processors per cluster must be a power of 2 no larger than 4. |
---|
28 | /////////////////////////////////////////////////////////////////////////////////////// |
---|
29 | |
---|
30 | #include "stdio.h" |
---|
31 | #include "stdlib.h" |
---|
32 | #include "user_barrier.h" |
---|
33 | #include "malloc.h" |
---|
34 | |
---|
35 | #define BLOCK_SIZE 512 // block size on disk |
---|
36 | #define X_MAX 16 // max number of clusters in row |
---|
37 | #define Y_MAX 16 // max number of clusters in column |
---|
38 | #define PROCS_MAX 4 // max number of procs per cluster |
---|
39 | #define CLUSTER_MAX (X_MAX * Y_MAX) // max number of clusters |
---|
40 | #define IMAGE_SIZE 256 // default image size |
---|
41 | #define INPUT_FILE_PATH "/misc/lena_256.raw" // default input file pathname |
---|
42 | #define OUTPUT_FILE_PATH "/home/lena_transposed.raw" // default output file pathname |
---|
43 | |
---|
44 | // macro to use a shared TTY |
---|
45 | #define printf(...); { lock_acquire( &tty_lock ); \ |
---|
46 | giet_tty_printf(__VA_ARGS__); \ |
---|
47 | lock_release( &tty_lock ); } |
---|
48 | |
---|
49 | /////////////////////////////////////////////////////// |
---|
50 | // global variables stored in seg_data in cluster(0,0) |
---|
51 | /////////////////////////////////////////////////////// |
---|
52 | |
---|
53 | // instrumentation counters for each processor in each cluster |
---|
54 | unsigned int LOAD_START[X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
55 | unsigned int LOAD_END [X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
56 | unsigned int TRSP_START[X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
57 | unsigned int TRSP_END [X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
58 | unsigned int DISP_START[X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
59 | unsigned int DISP_END [X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
60 | unsigned int STOR_START[X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
61 | unsigned int STOR_END [X_MAX][Y_MAX][PROCS_MAX] = {{{ 0 }}}; |
---|
62 | |
---|
63 | // arrays of pointers on distributed buffers |
---|
64 | // one input buffer & one output buffer per cluster |
---|
65 | unsigned char* buf_in [CLUSTER_MAX]; |
---|
66 | unsigned char* buf_out[CLUSTER_MAX]; |
---|
67 | |
---|
68 | // checksum variables |
---|
69 | unsigned check_line_before[1024]; |
---|
70 | unsigned check_line_after[1024]; |
---|
71 | |
---|
72 | // lock protecting shared TTY |
---|
73 | user_lock_t tty_lock; |
---|
74 | |
---|
75 | // synchronisation barrier (all threads) |
---|
76 | giet_sqt_barrier_t barrier; |
---|
77 | |
---|
78 | // input & output files pathname and size |
---|
79 | char input_file_name[256]; |
---|
80 | char output_file_name[256]; |
---|
81 | unsigned int image_size; |
---|
82 | |
---|
83 | //////////////////////////////////////////// |
---|
84 | __attribute__ ((constructor)) void execute() |
---|
85 | //////////////////////////////////////////// |
---|
86 | { |
---|
87 | unsigned int l; // line index for loops |
---|
88 | unsigned int p; // pixel index for loops |
---|
89 | |
---|
90 | // get processor identifiers |
---|
91 | unsigned int x_id; // x cluster coordinate |
---|
92 | unsigned int y_id; // y cluster coordinate |
---|
93 | unsigned int p_id; // local processor index |
---|
94 | |
---|
95 | giet_proc_xyp( &x_id, &y_id, &p_id); |
---|
96 | |
---|
97 | // get & check plat-form parameters |
---|
98 | unsigned int x_size; // number of clusters in a row |
---|
99 | unsigned int y_size; // number of clusters in a column |
---|
100 | unsigned int nprocs; // number of processors per cluster |
---|
101 | |
---|
102 | giet_procs_number( &x_size , &y_size , &nprocs ); |
---|
103 | |
---|
104 | unsigned int nclusters = x_size * y_size; // number of clusters |
---|
105 | unsigned int nthreads = x_size * y_size * nprocs; // number of threads |
---|
106 | unsigned int npixels = image_size * image_size; // pixels per image |
---|
107 | int fd_in = 0; // initial file descriptor |
---|
108 | int fd_out = 0; // output file descriptor |
---|
109 | unsigned int cluster_id = (x_id * y_size) + y_id; // "continuous" index |
---|
110 | unsigned int thread_id = (cluster_id * nprocs) + p_id; // "continuous" thread index |
---|
111 | |
---|
112 | // parallel load of image: |
---|
113 | // allocate buf_in and buf_out distributed buffers (one buf_in & one buf_out per cluster). |
---|
114 | // open input and output files, and load the relevant lines in local buf_in. |
---|
115 | // only thread running on processor[x,y,0] does it. |
---|
116 | |
---|
117 | LOAD_START[x_id][y_id][p_id] = giet_proctime(); |
---|
118 | |
---|
119 | if ( p_id == 0 ) |
---|
120 | { |
---|
121 | buf_in[cluster_id] = remote_malloc( npixels/nclusters, x_id, y_id ); |
---|
122 | buf_out[cluster_id] = remote_malloc( npixels/nclusters, x_id, y_id ); |
---|
123 | |
---|
124 | if ( (x_id==0) && (y_id==0) ) |
---|
125 | { |
---|
126 | printf("\n[TRANSPOSE] Proc [%d,%d,%d] completes buffer allocation at cycle %d\n", |
---|
127 | x_id, y_id, p_id, giet_proctime() ); |
---|
128 | } |
---|
129 | |
---|
130 | // open input file |
---|
131 | fd_in = giet_fat_open( input_file_name , O_RDONLY ); // read_only |
---|
132 | if ( fd_in < 0 ) |
---|
133 | { |
---|
134 | printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot open file %s\n", |
---|
135 | x_id , y_id , p_id , input_file_name ); |
---|
136 | giet_pthread_exit(" open() failure"); |
---|
137 | } |
---|
138 | else if ( (x_id==0) && (y_id==0) ) |
---|
139 | { |
---|
140 | printf("\n[TRANSPOSE] Proc [0,0,0] open file %s / fd = %d\n", |
---|
141 | input_file_name , fd_in ); |
---|
142 | } |
---|
143 | |
---|
144 | // open output file |
---|
145 | fd_out = giet_fat_open( output_file_name , O_CREATE ); // create if required |
---|
146 | if ( fd_out < 0 ) |
---|
147 | { |
---|
148 | printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot open file %s\n", |
---|
149 | x_id , y_id , p_id , output_file_name ); |
---|
150 | giet_pthread_exit(" open() failure"); |
---|
151 | } |
---|
152 | else if ( (x_id==0) && (y_id==0) ) |
---|
153 | { |
---|
154 | printf("\n[TRANSPOSE] Proc [0,0,0] open file %s / fd = %d\n", |
---|
155 | output_file_name , fd_out ); |
---|
156 | } |
---|
157 | |
---|
158 | |
---|
159 | unsigned int offset = ((npixels*cluster_id)/nclusters); |
---|
160 | if ( giet_fat_lseek( fd_in, |
---|
161 | offset, |
---|
162 | SEEK_SET ) != offset ) |
---|
163 | { |
---|
164 | printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot seek fd = %d\n", |
---|
165 | x_id , y_id , p_id , fd_in ); |
---|
166 | giet_pthread_exit(" seek() failure"); |
---|
167 | } |
---|
168 | |
---|
169 | unsigned int pixels = npixels / nclusters; |
---|
170 | if ( giet_fat_read( fd_in, |
---|
171 | buf_in[cluster_id], |
---|
172 | pixels ) != pixels ) |
---|
173 | { |
---|
174 | printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot read fd = %d\n", |
---|
175 | x_id , y_id , p_id , fd_in ); |
---|
176 | giet_pthread_exit(" read() failure"); |
---|
177 | } |
---|
178 | |
---|
179 | if ( (x_id==0) && (y_id==0) ) |
---|
180 | { |
---|
181 | printf("\n[TRANSPOSE] Proc [%d,%d,%d] completes load at cycle %d\n", |
---|
182 | x_id, y_id, p_id, giet_proctime() ); |
---|
183 | } |
---|
184 | } |
---|
185 | |
---|
186 | LOAD_END[x_id][y_id][p_id] = giet_proctime(); |
---|
187 | |
---|
188 | ///////////////////////////// |
---|
189 | sqt_barrier_wait( &barrier ); |
---|
190 | ///////////////////////////// |
---|
191 | |
---|
192 | // parallel transpose from buf_in to buf_out |
---|
193 | // each thread makes the transposition for nlt lines (nlt = image_size/nthreads) |
---|
194 | // from line [thread_id*nlt] to line [(thread_id + 1)*nlt - 1] |
---|
195 | // (p,l) are the absolute pixel coordinates in the source image |
---|
196 | |
---|
197 | TRSP_START[x_id][y_id][p_id] = giet_proctime(); |
---|
198 | |
---|
199 | unsigned int nlt = image_size / nthreads; // number of lines per thread |
---|
200 | unsigned int nlc = image_size / nclusters; // number of lines per cluster |
---|
201 | |
---|
202 | unsigned int src_cluster; |
---|
203 | unsigned int src_index; |
---|
204 | unsigned int dst_cluster; |
---|
205 | unsigned int dst_index; |
---|
206 | |
---|
207 | unsigned char byte; |
---|
208 | |
---|
209 | unsigned int first = thread_id * nlt; // first line index for a given thread |
---|
210 | unsigned int last = first + nlt; // last line index for a given thread |
---|
211 | |
---|
212 | for ( l = first ; l < last ; l++ ) |
---|
213 | { |
---|
214 | check_line_before[l] = 0; |
---|
215 | |
---|
216 | // in each iteration we transfer one byte |
---|
217 | for ( p = 0 ; p < image_size ; p++ ) |
---|
218 | { |
---|
219 | // read one byte from local buf_in |
---|
220 | src_cluster = l / nlc; |
---|
221 | src_index = (l % nlc)*image_size + p; |
---|
222 | byte = buf_in[src_cluster][src_index]; |
---|
223 | |
---|
224 | // compute checksum |
---|
225 | check_line_before[l] = check_line_before[l] + byte; |
---|
226 | |
---|
227 | // write one byte to remote buf_out |
---|
228 | dst_cluster = p / nlc; |
---|
229 | dst_index = (p % nlc)*image_size + l; |
---|
230 | buf_out[dst_cluster][dst_index] = byte; |
---|
231 | } |
---|
232 | } |
---|
233 | |
---|
234 | if ( (p_id == 0) && (x_id==0) && (y_id==0) ) |
---|
235 | { |
---|
236 | printf("\n[TRANSPOSE] proc [%d,%d,%d] completes transpose at cycle %d\n", |
---|
237 | x_id, y_id, p_id, giet_proctime() ); |
---|
238 | } |
---|
239 | |
---|
240 | TRSP_END[x_id][y_id][p_id] = giet_proctime(); |
---|
241 | |
---|
242 | ///////////////////////////// |
---|
243 | sqt_barrier_wait( &barrier ); |
---|
244 | ///////////////////////////// |
---|
245 | |
---|
246 | // parallel display from local buf_out to frame buffer |
---|
247 | // all threads contribute to display using memcpy... |
---|
248 | |
---|
249 | DISP_START[x_id][y_id][p_id] = giet_proctime(); |
---|
250 | |
---|
251 | unsigned int npt = npixels / nthreads; // number of pixels per thread |
---|
252 | |
---|
253 | giet_fbf_sync_write( npt * thread_id, |
---|
254 | &buf_out[cluster_id][p_id*npt], |
---|
255 | npt ); |
---|
256 | |
---|
257 | if ( (x_id==0) && (y_id==0) && (p_id==0) ) |
---|
258 | { |
---|
259 | printf("\n[TRANSPOSE] Proc [%d,%d,%d] completes display at cycle %d\n", |
---|
260 | x_id, y_id, p_id, giet_proctime() ); |
---|
261 | } |
---|
262 | |
---|
263 | DISP_END[x_id][y_id][p_id] = giet_proctime(); |
---|
264 | |
---|
265 | ///////////////////////////// |
---|
266 | sqt_barrier_wait( &barrier ); |
---|
267 | ///////////////////////////// |
---|
268 | |
---|
269 | // parallel store : buf_out buffers to disk |
---|
270 | // only thread running on processor(x,y,0) does it |
---|
271 | |
---|
272 | STOR_START[x_id][y_id][p_id] = giet_proctime(); |
---|
273 | |
---|
274 | if ( p_id == 0 ) |
---|
275 | { |
---|
276 | unsigned int offset = ((npixels*cluster_id)/nclusters); |
---|
277 | if ( giet_fat_lseek( fd_out, |
---|
278 | offset, |
---|
279 | SEEK_SET ) != offset ) |
---|
280 | { |
---|
281 | printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot seek fr = %d\n", |
---|
282 | x_id , y_id , p_id , fd_out ); |
---|
283 | giet_pthread_exit(" seek() failure"); |
---|
284 | } |
---|
285 | |
---|
286 | unsigned int pixels = npixels / nclusters; |
---|
287 | if ( giet_fat_write( fd_out, |
---|
288 | buf_out[cluster_id], |
---|
289 | pixels ) != pixels ) |
---|
290 | { |
---|
291 | printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot write fd = %d\n", |
---|
292 | x_id , y_id , p_id , fd_out ); |
---|
293 | giet_pthread_exit(" write() failure"); |
---|
294 | } |
---|
295 | |
---|
296 | if ( (x_id==0) && (y_id==0) ) |
---|
297 | { |
---|
298 | printf("\n[TRANSPOSE] Proc [%d,%d,%d] completes store at cycle %d\n", |
---|
299 | x_id, y_id, p_id, giet_proctime() ); |
---|
300 | } |
---|
301 | } |
---|
302 | |
---|
303 | STOR_END[x_id][y_id][p_id] = giet_proctime(); |
---|
304 | |
---|
305 | // In each cluster, only thread running on Processor[x,y,0] releases |
---|
306 | // the distributed buffers and close the file descriptors. |
---|
307 | |
---|
308 | if ( p_id==0 ) |
---|
309 | { |
---|
310 | free( buf_in[cluster_id] ); |
---|
311 | free( buf_out[cluster_id] ); |
---|
312 | |
---|
313 | giet_fat_close( fd_in ); |
---|
314 | giet_fat_close( fd_out ); |
---|
315 | } |
---|
316 | |
---|
317 | if ( (x_id != 0) || (y_id != 0) || (p_id != 0) ) |
---|
318 | giet_pthread_exit( "completed" ); |
---|
319 | |
---|
320 | } // end execute() |
---|
321 | |
---|
322 | |
---|
323 | |
---|
324 | ////////////////////////////////////// |
---|
325 | void instrument( unsigned int x_size, |
---|
326 | unsigned int y_size, |
---|
327 | unsigned int nprocs ) |
---|
328 | ////////////////////////////////////// |
---|
329 | { |
---|
330 | unsigned int x, y, l; |
---|
331 | |
---|
332 | unsigned int min_load_start = 0xFFFFFFFF; |
---|
333 | unsigned int max_load_start = 0; |
---|
334 | unsigned int min_load_ended = 0xFFFFFFFF; |
---|
335 | unsigned int max_load_ended = 0; |
---|
336 | unsigned int min_trsp_start = 0xFFFFFFFF; |
---|
337 | unsigned int max_trsp_start = 0; |
---|
338 | unsigned int min_trsp_ended = 0xFFFFFFFF; |
---|
339 | unsigned int max_trsp_ended = 0; |
---|
340 | unsigned int min_disp_start = 0xFFFFFFFF; |
---|
341 | unsigned int max_disp_start = 0; |
---|
342 | unsigned int min_disp_ended = 0xFFFFFFFF; |
---|
343 | unsigned int max_disp_ended = 0; |
---|
344 | unsigned int min_stor_start = 0xFFFFFFFF; |
---|
345 | unsigned int max_stor_start = 0; |
---|
346 | unsigned int min_stor_ended = 0xFFFFFFFF; |
---|
347 | unsigned int max_stor_ended = 0; |
---|
348 | |
---|
349 | for (x = 0; x < x_size; x++) |
---|
350 | { |
---|
351 | for (y = 0; y < y_size; y++) |
---|
352 | { |
---|
353 | for ( l = 0 ; l < nprocs ; l++ ) |
---|
354 | { |
---|
355 | if (LOAD_START[x][y][l] < min_load_start) min_load_start = LOAD_START[x][y][l]; |
---|
356 | if (LOAD_START[x][y][l] > max_load_start) max_load_start = LOAD_START[x][y][l]; |
---|
357 | if (LOAD_END[x][y][l] < min_load_ended) min_load_ended = LOAD_END[x][y][l]; |
---|
358 | if (LOAD_END[x][y][l] > max_load_ended) max_load_ended = LOAD_END[x][y][l]; |
---|
359 | if (TRSP_START[x][y][l] < min_trsp_start) min_trsp_start = TRSP_START[x][y][l]; |
---|
360 | if (TRSP_START[x][y][l] > max_trsp_start) max_trsp_start = TRSP_START[x][y][l]; |
---|
361 | if (TRSP_END[x][y][l] < min_trsp_ended) min_trsp_ended = TRSP_END[x][y][l]; |
---|
362 | if (TRSP_END[x][y][l] > max_trsp_ended) max_trsp_ended = TRSP_END[x][y][l]; |
---|
363 | if (DISP_START[x][y][l] < min_disp_start) min_disp_start = DISP_START[x][y][l]; |
---|
364 | if (DISP_START[x][y][l] > max_disp_start) max_disp_start = DISP_START[x][y][l]; |
---|
365 | if (DISP_END[x][y][l] < min_disp_ended) min_disp_ended = DISP_END[x][y][l]; |
---|
366 | if (DISP_END[x][y][l] > max_disp_ended) max_disp_ended = DISP_END[x][y][l]; |
---|
367 | if (STOR_START[x][y][l] < min_stor_start) min_stor_start = STOR_START[x][y][l]; |
---|
368 | if (STOR_START[x][y][l] > max_stor_start) max_stor_start = STOR_START[x][y][l]; |
---|
369 | if (STOR_END[x][y][l] < min_stor_ended) min_stor_ended = STOR_END[x][y][l]; |
---|
370 | if (STOR_END[x][y][l] > max_stor_ended) max_stor_ended = STOR_END[x][y][l]; |
---|
371 | } |
---|
372 | } |
---|
373 | } |
---|
374 | |
---|
375 | printf("\n ---------------- Instrumentation Results ---------------------\n"); |
---|
376 | |
---|
377 | printf(" - LOAD_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
378 | min_load_start, max_load_start, (min_load_start+max_load_start)/2, |
---|
379 | max_load_start-min_load_start); |
---|
380 | |
---|
381 | printf(" - LOAD_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
382 | min_load_ended, max_load_ended, (min_load_ended+max_load_ended)/2, |
---|
383 | max_load_ended-min_load_ended); |
---|
384 | |
---|
385 | printf(" - TRSP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
386 | min_trsp_start, max_trsp_start, (min_trsp_start+max_trsp_start)/2, |
---|
387 | max_trsp_start-min_trsp_start); |
---|
388 | |
---|
389 | printf(" - TRSP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
390 | min_trsp_ended, max_trsp_ended, (min_trsp_ended+max_trsp_ended)/2, |
---|
391 | max_trsp_ended-min_trsp_ended); |
---|
392 | |
---|
393 | printf(" - DISP_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
394 | min_disp_start, max_disp_start, (min_disp_start+max_disp_start)/2, |
---|
395 | max_disp_start-min_disp_start); |
---|
396 | |
---|
397 | printf(" - DISP_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
398 | min_disp_ended, max_disp_ended, (min_disp_ended+max_disp_ended)/2, |
---|
399 | max_disp_ended-min_disp_ended); |
---|
400 | |
---|
401 | printf(" - STOR_START : min = %d / max = %d / med = %d / delta = %d\n", |
---|
402 | min_stor_start, max_stor_start, (min_stor_start+max_stor_start)/2, |
---|
403 | max_stor_start-min_stor_start); |
---|
404 | |
---|
405 | printf(" - STOR_END : min = %d / max = %d / med = %d / delta = %d\n", |
---|
406 | min_stor_ended, max_stor_ended, (min_stor_ended+max_stor_ended)/2, |
---|
407 | max_stor_ended-min_stor_ended); |
---|
408 | |
---|
409 | } // end instrument() |
---|
410 | |
---|
411 | |
---|
412 | |
---|
413 | ////////////////////////////////////////// |
---|
414 | __attribute__ ((constructor)) void main() |
---|
415 | ////////////////////////////////////////// |
---|
416 | { |
---|
417 | // indexes for loops |
---|
418 | unsigned int x , y , n; |
---|
419 | |
---|
420 | // get identifiers for proc executing main |
---|
421 | unsigned int x_id; // x cluster coordinate |
---|
422 | unsigned int y_id; // y cluster coordinate |
---|
423 | unsigned int p_id; // local processor index |
---|
424 | |
---|
425 | giet_proc_xyp( &x_id , &y_id , &p_id ); |
---|
426 | |
---|
427 | // get & check plat-form parameters |
---|
428 | unsigned int x_size; // number of clusters in a row |
---|
429 | unsigned int y_size; // number of clusters in a column |
---|
430 | unsigned int nprocs; // number of processors per cluster |
---|
431 | |
---|
432 | giet_procs_number( &x_size , &y_size , &nprocs ); |
---|
433 | |
---|
434 | giet_pthread_assert( ((nprocs == 1) || (nprocs == 2) || (nprocs == 4)), |
---|
435 | "[TRANSPOSE ERROR] number of procs per cluster must be 1, 2 or 4"); |
---|
436 | |
---|
437 | giet_pthread_assert( ((x_size == 1) || (x_size == 2) || (x_size == 4) || |
---|
438 | (x_size == 8) || (x_size == 16)), |
---|
439 | "[TRANSPOSE ERROR] x_size must be 1,2,4,8,16"); |
---|
440 | |
---|
441 | giet_pthread_assert( ((y_size == 1) || (y_size == 2) || (y_size == 4) || |
---|
442 | (y_size == 8) || (y_size == 16)), |
---|
443 | "[TRANSPOSE ERROR] y_size must be 1,2,4,8,16"); |
---|
444 | |
---|
445 | // compute number of threads |
---|
446 | unsigned int nthreads = x_size * y_size * nprocs; |
---|
447 | |
---|
448 | // shared TTY allocation |
---|
449 | giet_tty_alloc( 1 ); |
---|
450 | lock_init( &tty_lock); |
---|
451 | |
---|
452 | // get FBF ownership and FBF size |
---|
453 | unsigned int width; |
---|
454 | unsigned int height; |
---|
455 | giet_fbf_alloc(); |
---|
456 | giet_fbf_size( &width , &height ); |
---|
457 | |
---|
458 | // enter interactive part if required |
---|
459 | printf("\n[TRANSPOSE] start at cycle %d on %d cores / FBF = %d * %d pixels\n", |
---|
460 | giet_proctime(), nthreads , width , height ); |
---|
461 | |
---|
462 | // input_file_name, output_file_name, and size acquisition |
---|
463 | printf("\n[TRANSPOSE] enter path for input file / default is : %s\n> ", INPUT_FILE_PATH ); |
---|
464 | giet_tty_gets( input_file_name , 256 ); |
---|
465 | printf("\n"); |
---|
466 | |
---|
467 | if ( strcmp( input_file_name , "" ) == 0 ) strcpy( input_file_name , INPUT_FILE_PATH ); |
---|
468 | |
---|
469 | printf("\n[TRANSPOSE] enter path for output file / default is : %s\n> ", OUTPUT_FILE_PATH ); |
---|
470 | giet_tty_gets( output_file_name , 256 ); |
---|
471 | printf("\n"); |
---|
472 | |
---|
473 | if ( strcmp( output_file_name , "" ) == 0 ) strcpy( output_file_name , OUTPUT_FILE_PATH ); |
---|
474 | |
---|
475 | printf("\n[TRANSPOSE] enter image size / default is : %d\n> ", IMAGE_SIZE ); |
---|
476 | giet_tty_getw( &image_size ); |
---|
477 | printf("\n"); |
---|
478 | |
---|
479 | if ( image_size == 0 ) image_size = IMAGE_SIZE; |
---|
480 | |
---|
481 | printf("\n[TRANSPOSE] input = %s / output = %s / size = %d\n", |
---|
482 | input_file_name, output_file_name, image_size ); |
---|
483 | |
---|
484 | giet_pthread_assert( (nprocs * x_size * y_size <= image_size ), |
---|
485 | "[TRANSPOSE ERROR] number of threads larger than number of lines"); |
---|
486 | |
---|
487 | // distributed heap initialisation |
---|
488 | for ( x = 0 ; x < x_size ; x++ ) |
---|
489 | { |
---|
490 | for ( y = 0 ; y < y_size ; y++ ) |
---|
491 | { |
---|
492 | heap_init( x , y ); |
---|
493 | } |
---|
494 | } |
---|
495 | |
---|
496 | // allocate thread[] array |
---|
497 | pthread_t* thread = malloc( nthreads * sizeof(pthread_t) ); |
---|
498 | |
---|
499 | // barrier initialisation |
---|
500 | sqt_barrier_init( &barrier, x_size , y_size , nprocs ); |
---|
501 | |
---|
502 | // Initialisation completed |
---|
503 | printf("\n[TRANSPOSE] initialisation completed at cycle %d\n", giet_proctime() ); |
---|
504 | |
---|
505 | // launch other threads to run execute() function |
---|
506 | for ( n = 1 ; n < nthreads ; n++ ) |
---|
507 | { |
---|
508 | if ( giet_pthread_create( &thread[n], |
---|
509 | NULL, // no attribute |
---|
510 | &execute, |
---|
511 | NULL ) ) // no argument |
---|
512 | { |
---|
513 | printf("\n[TRANSPOSE ERROR] creating thread %x\n", thread[n] ); |
---|
514 | giet_pthread_exit( NULL ); |
---|
515 | } |
---|
516 | } |
---|
517 | |
---|
518 | // run the execute() function |
---|
519 | execute(); |
---|
520 | |
---|
521 | // wait other threads completion |
---|
522 | for ( n = 1 ; n < nthreads ; n++ ) |
---|
523 | { |
---|
524 | if ( giet_pthread_join( thread[n], NULL ) ) |
---|
525 | { |
---|
526 | printf("\n[TRANSPOSE ERROR] joining thread %x\n", thread[n] ); |
---|
527 | giet_pthread_exit( NULL ); |
---|
528 | } |
---|
529 | else |
---|
530 | { |
---|
531 | printf("\n[TRANSPOSE] thread %x joined at cycle %d\n", |
---|
532 | thread[n] , giet_proctime() ); |
---|
533 | } |
---|
534 | } |
---|
535 | |
---|
536 | // call the instrument() function |
---|
537 | instrument( x_size , y_size , nprocs ); |
---|
538 | |
---|
539 | giet_pthread_exit( "completed" ); |
---|
540 | |
---|
541 | } // end main() |
---|
542 | |
---|