1 | /* -*- c++ -*- |
---|
2 | * |
---|
3 | * File : dspin_router.cpp |
---|
4 | * Copyright (c) UPMC, Lip6 |
---|
5 | * Authors : Alain Greiner, Abbas Sheibanyrad, Ivan Miro, Zhen Zhang |
---|
6 | * |
---|
7 | * SOCLIB_LGPL_HEADER_BEGIN |
---|
8 | * |
---|
9 | * This file is part of SoCLib, GNU LGPLv2.1. |
---|
10 | * |
---|
11 | * SoCLib is free software; you can redistribute it and/or modify it |
---|
12 | * under the terms of the GNU Lesser General Public License as published |
---|
13 | * by the Free Software Foundation; version 2.1 of the License. |
---|
14 | * |
---|
15 | * SoCLib is distributed in the hope that it will be useful, but |
---|
16 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
---|
18 | * Lesser General Public License for more details. |
---|
19 | * |
---|
20 | * You should have received a copy of the GNU Lesser General Public |
---|
21 | * License along with SoCLib; if not, write to the Free Software |
---|
22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
---|
23 | * 02110-1301 USA |
---|
24 | * |
---|
25 | * SOCLIB_LGPL_HEADER_END |
---|
26 | * |
---|
27 | */ |
---|
28 | |
---|
29 | /////////////////////////////////////////////////////////////////////////// |
---|
30 | // Implementation Note : |
---|
31 | // The xfirst_route(), broadcast_route() and is_broadcast() functions |
---|
32 | // defined below are used to decode the DSPIN first flit format: |
---|
33 | // - In case of a non-broadcast packet : |
---|
34 | // | X | Y |---------------------------------------|BC | |
---|
35 | // | x_width | y_width | flit_width - (x_width + y_width + 2) | 0 | |
---|
36 | // |
---|
37 | // - In case of a broacast |
---|
38 | // | XMIN | XMAX | YMIN | YMAX |-------------------|BC | |
---|
39 | // | 5 | 5 | 5 | 5 | flit_width - 22 | 1 | |
---|
40 | /////////////////////////////////////////////////////////////////////////// |
---|
41 | |
---|
42 | #include "../include/dspin_router.h" |
---|
43 | #include "dspin_router_config.h" |
---|
44 | |
---|
45 | namespace soclib { namespace caba { |
---|
46 | |
---|
47 | using namespace soclib::common; |
---|
48 | using namespace soclib::caba; |
---|
49 | |
---|
50 | #define tmpl(x) template<int flit_width> x DspinRouter<flit_width> |
---|
51 | |
---|
52 | //////////////////////////////////////////////// |
---|
53 | // constructor |
---|
54 | //////////////////////////////////////////////// |
---|
55 | tmpl(/**/)::DspinRouter( sc_module_name name, |
---|
56 | const size_t x, |
---|
57 | const size_t y, |
---|
58 | const size_t x_width, |
---|
59 | const size_t y_width, |
---|
60 | const size_t in_fifo_depth, |
---|
61 | const size_t out_fifo_depth, |
---|
62 | const bool broadcast_supported) |
---|
63 | : soclib::caba::BaseModule(name), |
---|
64 | |
---|
65 | p_clk( "p_clk" ), |
---|
66 | p_resetn( "p_resetn" ), |
---|
67 | p_in( alloc_elems<DspinInput<flit_width> >("p_in", 5) ), |
---|
68 | p_out( alloc_elems<DspinOutput<flit_width> >("p_out", 5) ), |
---|
69 | |
---|
70 | r_alloc_out( alloc_elems<sc_signal<bool> >("r_alloc_out", 5)), |
---|
71 | r_index_out( soclib::common::alloc_elems<sc_signal<size_t> >("r_index_out", 5)), |
---|
72 | r_fsm_in( alloc_elems<sc_signal<int> >("r_fsm_in", 5)), |
---|
73 | r_index_in( alloc_elems<sc_signal<size_t> >("r_index_in", 5)), |
---|
74 | |
---|
75 | m_local_x( x ), |
---|
76 | m_local_y( y ), |
---|
77 | m_x_width( x_width ), |
---|
78 | m_x_shift( flit_width - x_width ), |
---|
79 | m_x_mask( (0x1 << x_width) - 1 ), |
---|
80 | m_y_width( y_width ), |
---|
81 | m_y_shift( flit_width - x_width - y_width ), |
---|
82 | m_y_mask( (0x1 << y_width) - 1 ), |
---|
83 | m_broadcast_supported( broadcast_supported ), |
---|
84 | m_disable_mask( 0 ) |
---|
85 | { |
---|
86 | std::cout << " - Building DspinRouter : " << name << std::endl; |
---|
87 | |
---|
88 | SC_METHOD (transition); |
---|
89 | dont_initialize(); |
---|
90 | sensitive << p_clk.pos(); |
---|
91 | |
---|
92 | SC_METHOD (genMoore); |
---|
93 | dont_initialize(); |
---|
94 | sensitive << p_clk.neg(); |
---|
95 | |
---|
96 | r_fifo_in = (GenericFifo<internal_flit_t>*) |
---|
97 | malloc(sizeof(GenericFifo<internal_flit_t>) * 5); |
---|
98 | |
---|
99 | r_fifo_out = (GenericFifo<internal_flit_t>*) |
---|
100 | malloc(sizeof(GenericFifo<internal_flit_t>) * 5); |
---|
101 | |
---|
102 | r_buf_in = (internal_flit_t*) |
---|
103 | malloc(sizeof(internal_flit_t) * 5); |
---|
104 | |
---|
105 | for( size_t i = 0 ; i < 5 ; i++ ) |
---|
106 | { |
---|
107 | std::ostringstream stri; |
---|
108 | stri << "r_in_fifo_" << i; |
---|
109 | new(&r_fifo_in[i]) |
---|
110 | GenericFifo<internal_flit_t >(stri.str(), in_fifo_depth); |
---|
111 | |
---|
112 | std::ostringstream stro; |
---|
113 | stro << "r_out_fifo_" << i; |
---|
114 | new(&r_fifo_out[i]) |
---|
115 | GenericFifo<internal_flit_t >(stro.str(), out_fifo_depth); |
---|
116 | } |
---|
117 | |
---|
118 | p_recovery_cfg = NULL; |
---|
119 | } // end constructor |
---|
120 | |
---|
121 | /////////////////////////////////////////////////// |
---|
122 | tmpl(/**/)::~DspinRouter() |
---|
123 | { |
---|
124 | if ( p_recovery_cfg != NULL ) delete p_recovery_cfg; |
---|
125 | } |
---|
126 | |
---|
127 | /////////////////////////////////////////////////// |
---|
128 | tmpl(int)::xfirst_route( size_t xdest, size_t ydest ) |
---|
129 | { |
---|
130 | return (xdest < m_local_x ? REQ_WEST : |
---|
131 | (xdest > m_local_x ? REQ_EAST : |
---|
132 | (ydest < m_local_y ? REQ_SOUTH : |
---|
133 | (ydest > m_local_y ? REQ_NORTH : REQ_LOCAL)))); |
---|
134 | } |
---|
135 | |
---|
136 | /////////////////////////////////////////////////// |
---|
137 | tmpl(void)::bind_recovery_port( sc_core::sc_signal<uint32_t> &s ) |
---|
138 | { |
---|
139 | if (p_recovery_cfg == NULL) { |
---|
140 | p_recovery_cfg = new sc_core::sc_in<uint32_t> ("p_recovery_cfg"); |
---|
141 | } |
---|
142 | (*p_recovery_cfg)(s); |
---|
143 | } |
---|
144 | |
---|
145 | /////////////////////////////////////////////////// |
---|
146 | tmpl(bool)::is_destination_blackhole( size_t xdest, size_t ydest, int bhpos ) |
---|
147 | { |
---|
148 | size_t xhole, yhole; |
---|
149 | if (bhpos == BH_N) { |
---|
150 | xhole = m_local_x; |
---|
151 | yhole = m_local_y - 1; |
---|
152 | } |
---|
153 | else if (bhpos == BH_NW) { |
---|
154 | xhole = m_local_x + 1; |
---|
155 | yhole = m_local_y - 1; |
---|
156 | } |
---|
157 | else if (bhpos == BH_W) { |
---|
158 | xhole = m_local_x + 1; |
---|
159 | yhole = m_local_y; |
---|
160 | } |
---|
161 | else if (bhpos == BH_SW) { |
---|
162 | xhole = m_local_x + 1; |
---|
163 | yhole = m_local_y + 1; |
---|
164 | } |
---|
165 | else if (bhpos == BH_S) { |
---|
166 | xhole = m_local_x; |
---|
167 | yhole = m_local_y + 1; |
---|
168 | } |
---|
169 | else if (bhpos == BH_SE) { |
---|
170 | xhole = m_local_x - 1; |
---|
171 | yhole = m_local_y + 1; |
---|
172 | } |
---|
173 | else if (bhpos == BH_E) { |
---|
174 | xhole = m_local_x - 1; |
---|
175 | yhole = m_local_y; |
---|
176 | } |
---|
177 | else if (bhpos == BH_NE) { |
---|
178 | xhole = m_local_x - 1; |
---|
179 | yhole = m_local_y - 1; |
---|
180 | } |
---|
181 | else { |
---|
182 | return false; |
---|
183 | } |
---|
184 | |
---|
185 | return ((xdest == xhole) && (ydest == yhole)); |
---|
186 | } |
---|
187 | |
---|
188 | /////////////////////////////////////////////////// |
---|
189 | tmpl(int)::recovery_route( size_t xdest, size_t ydest ) |
---|
190 | { |
---|
191 | int bhpos = blackhole_position(); |
---|
192 | |
---|
193 | if ( xdest > m_local_x ) { |
---|
194 | if ( (bhpos == BH_NE) || (bhpos == BH_E) || (bhpos == BH_SE) || |
---|
195 | (bhpos == BH_S) ) { |
---|
196 | return REQ_EAST; |
---|
197 | } |
---|
198 | else if ( bhpos == BH_N ) { |
---|
199 | if ( (m_local_y == 1) || (m_local_x == 0) || (ydest >= m_local_y) || |
---|
200 | (xdest > (m_local_x + 1)) ) { |
---|
201 | return REQ_EAST; |
---|
202 | } |
---|
203 | else { |
---|
204 | return REQ_WEST; |
---|
205 | } |
---|
206 | } |
---|
207 | else if ( bhpos == BH_NW ) { |
---|
208 | if ( (m_local_y == 1) || (ydest >= m_local_y) || |
---|
209 | (xdest > (m_local_x + 2)) ) { |
---|
210 | return REQ_EAST; |
---|
211 | } |
---|
212 | else { |
---|
213 | return REQ_SOUTH; |
---|
214 | } |
---|
215 | } |
---|
216 | else if ( bhpos == BH_W ) { |
---|
217 | if ( (m_local_y == 0) || (ydest > m_local_y)) { |
---|
218 | return REQ_NORTH; |
---|
219 | } |
---|
220 | else { |
---|
221 | return REQ_SOUTH; |
---|
222 | } |
---|
223 | } |
---|
224 | else if ( bhpos == BH_SW ) { |
---|
225 | if ( (ydest <= m_local_y) || (xdest > (m_local_x + 1)) ) { |
---|
226 | return REQ_EAST; |
---|
227 | } |
---|
228 | else { |
---|
229 | return REQ_NORTH; |
---|
230 | } |
---|
231 | } |
---|
232 | std::cout << "error: unexpected condition in function " |
---|
233 | << __FILE__ << ":" << __func__ << " +" << __LINE__ |
---|
234 | << std::endl; |
---|
235 | exit(1); |
---|
236 | } // end if (xdest > m_local_x) |
---|
237 | else if ( xdest < m_local_x ) { |
---|
238 | if ( (bhpos == BH_N) || (bhpos == BH_NW) || (bhpos == BH_W) || |
---|
239 | (bhpos == BH_SW) || (bhpos == BH_S) ) { |
---|
240 | return REQ_WEST; |
---|
241 | } |
---|
242 | else if ( bhpos == BH_NE ) { |
---|
243 | if ( (xdest < (m_local_x - 1)) || (ydest >= m_local_y) ) { |
---|
244 | return REQ_WEST; |
---|
245 | } |
---|
246 | else { |
---|
247 | return REQ_SOUTH; |
---|
248 | } |
---|
249 | } |
---|
250 | else if ( bhpos == BH_SE ) { |
---|
251 | if ( (m_local_x == 1) && (ydest > (m_local_y + 1)) ) { |
---|
252 | return REQ_NORTH; |
---|
253 | } |
---|
254 | else { |
---|
255 | return REQ_WEST; |
---|
256 | } |
---|
257 | } |
---|
258 | else if ( bhpos == BH_E ) { |
---|
259 | if ( (m_local_y == 0) || |
---|
260 | ((m_local_x == 1) && (ydest > m_local_y)) ) { |
---|
261 | return REQ_NORTH; |
---|
262 | } |
---|
263 | else { |
---|
264 | return REQ_SOUTH; |
---|
265 | } |
---|
266 | } |
---|
267 | std::cout << "error: unexpected condition in function " |
---|
268 | << __FILE__ << ":" << __func__ << " +" << __LINE__ |
---|
269 | << std::endl; |
---|
270 | exit(1); |
---|
271 | } // end if (xdest < m_local_x) |
---|
272 | else if ( ydest > m_local_y ) { |
---|
273 | if ( bhpos != BH_S ) { |
---|
274 | return REQ_NORTH; |
---|
275 | } |
---|
276 | else if ( m_local_x != 0 ) { |
---|
277 | return REQ_WEST; |
---|
278 | } |
---|
279 | else { |
---|
280 | return REQ_EAST; |
---|
281 | } |
---|
282 | } // end if (ydest > m_local_y) |
---|
283 | else if ( ydest < m_local_y ) { |
---|
284 | if ( bhpos != BH_N ) { |
---|
285 | return REQ_SOUTH; |
---|
286 | } |
---|
287 | else if ( m_local_x != 0) { |
---|
288 | return REQ_WEST; |
---|
289 | } |
---|
290 | else { |
---|
291 | return REQ_EAST; |
---|
292 | } |
---|
293 | } // end if (ydest < m_local_y) |
---|
294 | return REQ_LOCAL; |
---|
295 | } |
---|
296 | |
---|
297 | /////////////////////////////////////////////////// |
---|
298 | tmpl(int)::route( sc_uint<flit_width> data ) |
---|
299 | { |
---|
300 | size_t xdest = (size_t)(data >> m_x_shift) & m_x_mask; |
---|
301 | size_t ydest = (size_t)(data >> m_y_shift) & m_y_mask; |
---|
302 | if ( p_recovery_cfg != NULL ) |
---|
303 | { |
---|
304 | if (blackhole_position() != BH_NONE ) |
---|
305 | { |
---|
306 | // reroute the request if its destination is the blackhole (this |
---|
307 | // is to implement the segment recovery mechanism) |
---|
308 | if (is_destination_blackhole(xdest, ydest, blackhole_position())) |
---|
309 | { |
---|
310 | int dir = migration_route(); |
---|
311 | |
---|
312 | #if SOCLIB_MODULE_DEBUG |
---|
313 | std::cout << "<" << name() << "> migration: " |
---|
314 | << "route request to DIR = " << dir << std::endl; |
---|
315 | #endif |
---|
316 | return dir; |
---|
317 | } |
---|
318 | |
---|
319 | if (is_network_recovery_enable()) |
---|
320 | { |
---|
321 | int dir = recovery_route(xdest, ydest); |
---|
322 | |
---|
323 | #if SOCLIB_MODULE_DEBUG |
---|
324 | std::cout << "<" << name() << "> network recovery: " |
---|
325 | << "route request to DIR = " << dir << std::endl; |
---|
326 | #endif |
---|
327 | return dir; |
---|
328 | } |
---|
329 | } |
---|
330 | } |
---|
331 | return xfirst_route(xdest, ydest); |
---|
332 | } |
---|
333 | |
---|
334 | ////////////////////////////////////////////////////////////////////////// |
---|
335 | tmpl(int)::broadcast_route(int step, int source, sc_uint<flit_width> data) |
---|
336 | { |
---|
337 | int sel = REQ_NOP; |
---|
338 | size_t xmin = (data >> (flit_width - 5 )) & 0x1F; |
---|
339 | size_t xmax = (data >> (flit_width - 10)) & 0x1F; |
---|
340 | size_t ymin = (data >> (flit_width - 15)) & 0x1F; |
---|
341 | size_t ymax = (data >> (flit_width - 20)) & 0x1F; |
---|
342 | |
---|
343 | switch(source) { |
---|
344 | case REQ_LOCAL : |
---|
345 | if ( step == 1 ) sel = REQ_NORTH; |
---|
346 | else if ( step == 2 ) sel = REQ_SOUTH; |
---|
347 | else if ( step == 3 ) sel = REQ_EAST; |
---|
348 | else if ( step == 4 ) sel = REQ_WEST; |
---|
349 | break; |
---|
350 | case REQ_NORTH : |
---|
351 | if ( step == 1 ) sel = REQ_SOUTH; |
---|
352 | else if ( step == 2 ) sel = REQ_LOCAL; |
---|
353 | else if ( step == 3 ) sel = REQ_NOP; |
---|
354 | else if ( step == 4 ) sel = REQ_NOP; |
---|
355 | break; |
---|
356 | case REQ_SOUTH : |
---|
357 | if ( step == 1 ) sel = REQ_NORTH; |
---|
358 | else if ( step == 2 ) sel = REQ_LOCAL; |
---|
359 | else if ( step == 3 ) sel = REQ_NOP; |
---|
360 | else if ( step == 4 ) sel = REQ_NOP; |
---|
361 | break; |
---|
362 | case REQ_EAST : |
---|
363 | if ( step == 1 ) sel = REQ_WEST; |
---|
364 | else if ( step == 2 ) sel = REQ_NORTH; |
---|
365 | else if ( step == 3 ) sel = REQ_SOUTH; |
---|
366 | else if ( step == 4 ) sel = REQ_LOCAL; |
---|
367 | break; |
---|
368 | case REQ_WEST : |
---|
369 | if ( step == 1 ) sel = REQ_EAST; |
---|
370 | else if ( step == 2 ) sel = REQ_NORTH; |
---|
371 | else if ( step == 3 ) sel = REQ_SOUTH; |
---|
372 | else if ( step == 4 ) sel = REQ_LOCAL; |
---|
373 | break; |
---|
374 | } |
---|
375 | if ( (sel == REQ_NORTH) && !(m_local_y < ymax) ) sel = REQ_NOP; |
---|
376 | else if ( (sel == REQ_SOUTH) && !(m_local_y > ymin) ) sel = REQ_NOP; |
---|
377 | else if ( (sel == REQ_EAST ) && !(m_local_x < xmax) ) sel = REQ_NOP; |
---|
378 | else if ( (sel == REQ_WEST ) && !(m_local_x > xmin) ) sel = REQ_NOP; |
---|
379 | |
---|
380 | return sel; |
---|
381 | } |
---|
382 | |
---|
383 | ///////////////////////////////////////////////////////// |
---|
384 | tmpl(inline bool)::is_broadcast(sc_uint<flit_width> data) |
---|
385 | { |
---|
386 | return ( (data & 0x1) != 0); |
---|
387 | } |
---|
388 | |
---|
389 | ///////////////////////// |
---|
390 | tmpl(void)::print_trace() |
---|
391 | { |
---|
392 | const char* port_name[] = |
---|
393 | { |
---|
394 | "N", |
---|
395 | "S", |
---|
396 | "E", |
---|
397 | "W", |
---|
398 | "L" |
---|
399 | }; |
---|
400 | |
---|
401 | const char* infsm_str[] = |
---|
402 | { |
---|
403 | "IDLE", |
---|
404 | "REQ", |
---|
405 | "ALLOC", |
---|
406 | "REQ_FIRST", |
---|
407 | "ALLOC_FIRST", |
---|
408 | "REQ_SECOND", |
---|
409 | "ALLOC_SECOND", |
---|
410 | "REQ_THIRD", |
---|
411 | "ALLOC_THIRD", |
---|
412 | "REQ_FOURTH", |
---|
413 | "ALLOC_FOURTH" |
---|
414 | }; |
---|
415 | |
---|
416 | std::cout << "DSPIN_ROUTER " << name(); |
---|
417 | |
---|
418 | for( size_t i = 0 ; i < 5 ; i++) // loop on input ports |
---|
419 | { |
---|
420 | std::cout << " / infsm[" << port_name[i] << "] " |
---|
421 | << infsm_str[r_fsm_in[i].read()]; |
---|
422 | } |
---|
423 | |
---|
424 | for ( size_t out=0 ; out<5 ; out++) // loop on output ports |
---|
425 | { |
---|
426 | if ( r_alloc_out[out].read() ) |
---|
427 | { |
---|
428 | int in = r_index_out[out]; |
---|
429 | std::cout << " / " << port_name[in] << " -> " << port_name[out] ; |
---|
430 | } |
---|
431 | } |
---|
432 | std::cout << std::endl; |
---|
433 | } |
---|
434 | |
---|
435 | //////////////////////// |
---|
436 | tmpl(void)::transition() |
---|
437 | { |
---|
438 | // Long wires connecting input and output ports |
---|
439 | size_t req_in[5]; // input ports -> output ports |
---|
440 | size_t get_out[5]; // output ports -> input ports |
---|
441 | bool put_in[5]; // input ports -> output ports |
---|
442 | internal_flit_t data_in[5]; // input ports -> output ports |
---|
443 | |
---|
444 | // control signals for the input fifos |
---|
445 | bool fifo_in_write[5]; |
---|
446 | bool fifo_in_read[5]; |
---|
447 | internal_flit_t fifo_in_wdata[5]; |
---|
448 | |
---|
449 | // control signals for the output fifos |
---|
450 | bool fifo_out_write[5]; |
---|
451 | bool fifo_out_read[5]; |
---|
452 | internal_flit_t fifo_out_wdata[5]; |
---|
453 | |
---|
454 | // Reset |
---|
455 | if ( p_resetn == false ) |
---|
456 | { |
---|
457 | for(size_t i = 0 ; i < 5 ; i++) |
---|
458 | { |
---|
459 | r_alloc_out[i] = false; |
---|
460 | r_index_out[i] = 0; |
---|
461 | r_index_in[i] = 0; |
---|
462 | r_fsm_in[i] = INFSM_IDLE; |
---|
463 | r_fifo_in[i].init(); |
---|
464 | r_fifo_out[i].init(); |
---|
465 | } |
---|
466 | return; |
---|
467 | } |
---|
468 | |
---|
469 | // fifos signals default values |
---|
470 | for(size_t i = 0 ; i < 5 ; i++) |
---|
471 | { |
---|
472 | fifo_in_read[i] = false; |
---|
473 | |
---|
474 | // do not write into the FIFO of disabled interfaces |
---|
475 | fifo_in_write[i] = p_in[i].write.read() && |
---|
476 | ((m_disable_mask & (1 << i)) == 0); |
---|
477 | |
---|
478 | fifo_in_wdata[i].data = p_in[i].data.read(); |
---|
479 | fifo_in_wdata[i].eop = p_in[i].eop.read(); |
---|
480 | |
---|
481 | fifo_out_read[i] = p_out[i].read.read(); |
---|
482 | fifo_out_write[i] = false; |
---|
483 | } |
---|
484 | |
---|
485 | // loop on the output ports: |
---|
486 | // compute get_out[j] depending on the output port state |
---|
487 | // and combining fifo_out[j].wok and r_alloc_out[j] |
---|
488 | for ( size_t j = 0 ; j < 5 ; j++ ) |
---|
489 | { |
---|
490 | if( r_alloc_out[j].read() and (r_fifo_out[j].wok()) ) |
---|
491 | { |
---|
492 | get_out[j] = r_index_out[j].read(); |
---|
493 | } |
---|
494 | else |
---|
495 | { |
---|
496 | get_out[j] = 0xFFFFFFFF; |
---|
497 | } |
---|
498 | } |
---|
499 | |
---|
500 | // loop on the input ports : |
---|
501 | // The port state is defined by r_fsm_in[i], r_index_in[i] & r_buf_in[i] |
---|
502 | // The req_in[i] computation implements the X-FIRST algorithm. |
---|
503 | // data_in[i], put_in[i] and req_in[i] depend on the input port state. |
---|
504 | // The fifo_in_read[i] is computed further... |
---|
505 | |
---|
506 | for ( size_t i = 0 ; i < 5 ; i++ ) |
---|
507 | { |
---|
508 | switch ( r_fsm_in[i].read() ) |
---|
509 | { |
---|
510 | case INFSM_IDLE: // no output port allocated |
---|
511 | { |
---|
512 | put_in[i] = false; |
---|
513 | |
---|
514 | if ( r_fifo_in[i].rok() ) // packet available in input fifo |
---|
515 | { |
---|
516 | if ( is_broadcast( r_fifo_in[i].read().data ) and |
---|
517 | m_broadcast_supported ) // broadcast |
---|
518 | { |
---|
519 | fifo_in_read[i] = true; |
---|
520 | req_in[i] = broadcast_route(1, i, r_fifo_in[i].read().data); |
---|
521 | r_buf_in[i] = r_fifo_in[i].read(); |
---|
522 | r_index_in[i] = req_in[i]; |
---|
523 | if( req_in[i] == REQ_NOP ) r_fsm_in[i] = INFSM_REQ_SECOND; |
---|
524 | else r_fsm_in[i] = INFSM_REQ_FIRST; |
---|
525 | } |
---|
526 | else // unicast |
---|
527 | { |
---|
528 | req_in[i] = route(r_fifo_in[i].read().data); |
---|
529 | r_index_in[i] = req_in[i]; |
---|
530 | r_fsm_in[i] = INFSM_REQ; |
---|
531 | } |
---|
532 | } |
---|
533 | else |
---|
534 | { |
---|
535 | req_in[i] = REQ_NOP; |
---|
536 | } |
---|
537 | break; |
---|
538 | } |
---|
539 | case INFSM_REQ: // not a broadcast / waiting output port allocation |
---|
540 | { |
---|
541 | data_in[i] = r_fifo_in[i].read(); |
---|
542 | put_in[i] = r_fifo_in[i].rok(); |
---|
543 | req_in[i] = r_index_in[i]; |
---|
544 | fifo_in_read[i] = (get_out[r_index_in[i].read()] == i); |
---|
545 | if ( get_out[r_index_in[i].read()] == i ) // first flit transfered |
---|
546 | { |
---|
547 | if ( r_fifo_in[i].read().eop ) r_fsm_in[i] = INFSM_IDLE; |
---|
548 | else r_fsm_in[i] = INFSM_ALLOC; |
---|
549 | } |
---|
550 | break; |
---|
551 | } |
---|
552 | case INFSM_ALLOC: // not a broadcast / output port allocated |
---|
553 | { |
---|
554 | data_in[i] = r_fifo_in[i].read(); |
---|
555 | put_in[i] = r_fifo_in[i].rok(); |
---|
556 | req_in[i] = REQ_NOP; // no request |
---|
557 | fifo_in_read[i] = (get_out[r_index_in[i].read()] == i); |
---|
558 | if ( r_fifo_in[i].read().eop and |
---|
559 | r_fifo_in[i].rok() and |
---|
560 | (get_out[r_index_in[i].read()] == i) ) // last flit transfered |
---|
561 | { |
---|
562 | r_fsm_in[i] = INFSM_IDLE; |
---|
563 | } |
---|
564 | break; |
---|
565 | } |
---|
566 | case INFSM_REQ_FIRST: // broacast / waiting first output port allocation |
---|
567 | { |
---|
568 | data_in[i] = r_buf_in[i]; |
---|
569 | put_in[i] = true; |
---|
570 | req_in[i] = broadcast_route(1, i, r_buf_in[i].data); |
---|
571 | r_index_in[i] = req_in[i]; |
---|
572 | if ( req_in[i] == REQ_NOP ) // no transfer for this step |
---|
573 | { |
---|
574 | r_fsm_in[i] = INFSM_REQ_SECOND; |
---|
575 | } |
---|
576 | else |
---|
577 | { |
---|
578 | if( get_out[req_in[i]] == i ) // header flit transfered |
---|
579 | { |
---|
580 | r_fsm_in[i] = INFSM_ALLOC_FIRST; |
---|
581 | } |
---|
582 | } |
---|
583 | break; |
---|
584 | } |
---|
585 | case INFSM_ALLOC_FIRST: // broadcast / first output port allocated |
---|
586 | { |
---|
587 | data_in[i] = r_fifo_in[i].read(); |
---|
588 | put_in[i] = r_fifo_in[i].rok(); |
---|
589 | req_in[i] = REQ_NOP; |
---|
590 | if( (get_out[r_index_in[i].read()] == i) |
---|
591 | and r_fifo_in[i].rok() ) // data flit transfered |
---|
592 | { |
---|
593 | if ( not r_fifo_in[i].read().eop ) |
---|
594 | { |
---|
595 | std::cout << "ERROR in DSPIN_ROUTER " << name() |
---|
596 | << " : broadcast packet must be 2 flits" << std::endl; |
---|
597 | } |
---|
598 | r_fsm_in[i] = INFSM_REQ_SECOND; |
---|
599 | } |
---|
600 | break; |
---|
601 | } |
---|
602 | case INFSM_REQ_SECOND: // broacast / waiting second output port allocation |
---|
603 | { |
---|
604 | data_in[i] = r_buf_in[i]; |
---|
605 | put_in[i] = true; |
---|
606 | req_in[i] = broadcast_route(2, i, r_buf_in[i].data); |
---|
607 | r_index_in[i] = req_in[i]; |
---|
608 | if ( req_in[i] == REQ_NOP ) // no transfer for this step |
---|
609 | { |
---|
610 | r_fsm_in[i] = INFSM_REQ_THIRD; |
---|
611 | } |
---|
612 | else |
---|
613 | { |
---|
614 | if( get_out[req_in[i]] == i ) // header flit transfered |
---|
615 | { |
---|
616 | r_fsm_in[i] = INFSM_ALLOC_SECOND; |
---|
617 | } |
---|
618 | } |
---|
619 | break; |
---|
620 | } |
---|
621 | case INFSM_ALLOC_SECOND: // broadcast / second output port allocated |
---|
622 | { |
---|
623 | data_in[i] = r_fifo_in[i].read(); |
---|
624 | put_in[i] = r_fifo_in[i].rok(); |
---|
625 | req_in[i] = REQ_NOP; |
---|
626 | if( (get_out[r_index_in[i].read()] == i ) |
---|
627 | and r_fifo_in[i].rok() ) // data flit transfered |
---|
628 | { |
---|
629 | if ( not r_fifo_in[i].read().eop ) |
---|
630 | { |
---|
631 | std::cout << "ERROR in DSPIN_ROUTER " << name() |
---|
632 | << " : broadcast packet must be 2 flits" << std::endl; |
---|
633 | } |
---|
634 | r_fsm_in[i] = INFSM_REQ_THIRD; |
---|
635 | } |
---|
636 | break; |
---|
637 | } |
---|
638 | case INFSM_REQ_THIRD: // broacast / waiting third output port allocation |
---|
639 | { |
---|
640 | data_in[i] = r_buf_in[i]; |
---|
641 | put_in[i] = true; |
---|
642 | req_in[i] = broadcast_route(3, i, r_buf_in[i].data); |
---|
643 | r_index_in[i] = req_in[i]; |
---|
644 | if ( req_in[i] == REQ_NOP ) // no transfer for this step |
---|
645 | { |
---|
646 | r_fsm_in[i] = INFSM_REQ_FOURTH; |
---|
647 | } |
---|
648 | else |
---|
649 | { |
---|
650 | if( get_out[req_in[i]] == i ) // header flit transfered |
---|
651 | { |
---|
652 | r_fsm_in[i] = INFSM_ALLOC_THIRD; |
---|
653 | } |
---|
654 | } |
---|
655 | break; |
---|
656 | } |
---|
657 | case INFSM_ALLOC_THIRD: // broadcast / third output port allocated |
---|
658 | { |
---|
659 | data_in[i] = r_fifo_in[i].read(); |
---|
660 | put_in[i] = r_fifo_in[i].rok(); |
---|
661 | req_in[i] = REQ_NOP; |
---|
662 | if( (get_out[r_index_in[i].read()] == i ) |
---|
663 | and r_fifo_in[i].rok() ) // data flit transfered |
---|
664 | { |
---|
665 | if ( not r_fifo_in[i].read().eop ) |
---|
666 | { |
---|
667 | std::cout << "ERROR in DSPIN_ROUTER " << name() |
---|
668 | << " : broadcast packet must be 2 flits" << std::endl; |
---|
669 | } |
---|
670 | r_fsm_in[i] = INFSM_REQ_FOURTH; |
---|
671 | } |
---|
672 | break; |
---|
673 | } |
---|
674 | case INFSM_REQ_FOURTH: // broacast / waiting fourth output port allocation |
---|
675 | { |
---|
676 | data_in[i] = r_buf_in[i]; |
---|
677 | put_in[i] = true; |
---|
678 | req_in[i] = broadcast_route(4, i, r_buf_in[i].data); |
---|
679 | r_index_in[i] = req_in[i]; |
---|
680 | if ( req_in[i] == REQ_NOP ) // no transfer for this step |
---|
681 | { |
---|
682 | fifo_in_read[i] = true; |
---|
683 | r_fsm_in[i] = INFSM_IDLE; |
---|
684 | } |
---|
685 | else |
---|
686 | { |
---|
687 | if( get_out[req_in[i]] == i ) // header flit transfered |
---|
688 | { |
---|
689 | r_fsm_in[i] = INFSM_ALLOC_FOURTH; |
---|
690 | } |
---|
691 | } |
---|
692 | break; |
---|
693 | } |
---|
694 | case INFSM_ALLOC_FOURTH: // broadcast / fourth output port allocated |
---|
695 | { |
---|
696 | data_in[i] = r_fifo_in[i].read(); |
---|
697 | put_in[i] = r_fifo_in[i].rok(); |
---|
698 | req_in[i] = REQ_NOP; |
---|
699 | if( (get_out[r_index_in[i].read()] == i ) |
---|
700 | and r_fifo_in[i].rok() ) // data flit transfered |
---|
701 | { |
---|
702 | if ( not r_fifo_in[i].read().eop ) |
---|
703 | { |
---|
704 | std::cout << "ERROR in DSPIN_ROUTER " << name() |
---|
705 | << " : broadcast packet must be 2 flits" << std::endl; |
---|
706 | } |
---|
707 | fifo_in_read[i] = true; |
---|
708 | r_fsm_in[i] = INFSM_IDLE; |
---|
709 | } |
---|
710 | break; |
---|
711 | } |
---|
712 | } // end switch |
---|
713 | } // end for input ports |
---|
714 | |
---|
715 | // loop on the output ports : |
---|
716 | // The r_alloc_out[j] and r_index_out[j] computation |
---|
717 | // implements the round-robin allocation policy. |
---|
718 | // These two registers implement a 10 states FSM. |
---|
719 | for( size_t j = 0 ; j < 5 ; j++ ) |
---|
720 | { |
---|
721 | if( not r_alloc_out[j].read() ) // not allocated: possible new allocation |
---|
722 | { |
---|
723 | for( size_t k = r_index_out[j].read() + 1 ; |
---|
724 | k < (r_index_out[j] + 6) ; k++) |
---|
725 | { |
---|
726 | size_t i = k % 5; |
---|
727 | |
---|
728 | if( req_in[i] == j ) |
---|
729 | { |
---|
730 | r_alloc_out[j] = true; |
---|
731 | r_index_out[j] = i; |
---|
732 | break; |
---|
733 | } |
---|
734 | } // end loop on input ports |
---|
735 | } |
---|
736 | else // allocated: possible desallocation |
---|
737 | { |
---|
738 | if ( data_in[r_index_out[j]].eop and |
---|
739 | r_fifo_out[j].wok() and |
---|
740 | put_in[r_index_out[j]] ) |
---|
741 | { |
---|
742 | r_alloc_out[j] = false; |
---|
743 | } |
---|
744 | } |
---|
745 | } // end loop on output ports |
---|
746 | |
---|
747 | // loop on the output ports : |
---|
748 | // The fifo_out_write[j] and fifo_out_wdata[j] computation |
---|
749 | // implements the output port mux. |
---|
750 | for( size_t j = 0 ; j < 5 ; j++ ) |
---|
751 | { |
---|
752 | if( r_alloc_out[j] ) // output port allocated |
---|
753 | { |
---|
754 | fifo_out_write[j] = put_in[r_index_out[j]] && |
---|
755 | ((m_disable_mask & (1 << j)) == 0); |
---|
756 | fifo_out_wdata[j] = data_in[r_index_out[j]]; |
---|
757 | } |
---|
758 | } // end loop on the output ports |
---|
759 | |
---|
760 | // FIFOS update |
---|
761 | for(size_t i = 0 ; i < 5 ; i++) |
---|
762 | { |
---|
763 | r_fifo_in[i].update(fifo_in_read[i], |
---|
764 | fifo_in_write[i], |
---|
765 | fifo_in_wdata[i]); |
---|
766 | r_fifo_out[i].update(fifo_out_read[i], |
---|
767 | fifo_out_write[i], |
---|
768 | fifo_out_wdata[i]); |
---|
769 | } |
---|
770 | } // end transition |
---|
771 | |
---|
772 | //////////////////////////////// |
---|
773 | // genMoore |
---|
774 | //////////////////////////////// |
---|
775 | tmpl(void)::genMoore() |
---|
776 | { |
---|
777 | for(size_t i = 0 ; i < 5 ; i++) |
---|
778 | { |
---|
779 | // input ports : READ signals |
---|
780 | p_in[i].read = r_fifo_in[i].wok(); |
---|
781 | |
---|
782 | // output ports : DATA & WRITE signals |
---|
783 | p_out[i].data = r_fifo_out[i].read().data; |
---|
784 | p_out[i].eop = r_fifo_out[i].read().eop; |
---|
785 | p_out[i].write = r_fifo_out[i].rok(); |
---|
786 | } |
---|
787 | } // end genMoore |
---|
788 | |
---|
789 | }} // end namespace |
---|
790 | |
---|
791 | // Local Variables: |
---|
792 | // tab-width: 4 |
---|
793 | // c-basic-offset: 4 |
---|
794 | // c-file-offsets:((innamespace . 0)(inline-open . 0)) |
---|
795 | // indent-tabs-mode: nil |
---|
796 | // End: |
---|
797 | |
---|
798 | // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=4:softtabstop=4 |
---|