[549] | 1 | /* -*- c++ -*- |
---|
| 2 | * |
---|
| 3 | * File : dspin_router_tsar.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 | // This component implements a variant of the standard (SocLib) DSPIN router: |
---|
| 31 | // The routing function has been modified to handle the special case of |
---|
| 32 | // cluster_iob0 (containing component IOB0) and cluster_iob1 (containing |
---|
| 33 | // component IOB1). In those two cluster, the response router must decode |
---|
| 34 | // both the SRCID global bits AND the SRCID local bits to distinguish |
---|
| 35 | // between the IOB and MEMC initiators. |
---|
| 36 | // This component contains the following modifications: |
---|
[578] | 37 | // - 4 new constructor arguments |
---|
[549] | 38 | // - 6 new member variables |
---|
| 39 | // - a modified routing function |
---|
| 40 | //////////////////////////////////////////////////////////////////////////////// |
---|
| 41 | |
---|
| 42 | #include "../include/dspin_router_tsar.h" |
---|
| 43 | |
---|
| 44 | namespace soclib { namespace caba { |
---|
| 45 | |
---|
| 46 | using namespace soclib::common; |
---|
| 47 | using namespace soclib::caba; |
---|
| 48 | |
---|
| 49 | #define tmpl(x) template<int flit_width> x DspinRouterTsar<flit_width> |
---|
| 50 | |
---|
| 51 | //////////////////////////////////////////////// |
---|
| 52 | // constructor |
---|
| 53 | //////////////////////////////////////////////// |
---|
| 54 | tmpl(/**/)::DspinRouterTsar( |
---|
| 55 | sc_module_name name, |
---|
| 56 | const size_t x, // x coordinate |
---|
| 57 | const size_t y, // y cordinate |
---|
| 58 | const size_t x_width, // x field width in first flit |
---|
| 59 | const size_t y_width, // y field width in first flit |
---|
| 60 | const size_t in_fifo_depth, // input fifo depth |
---|
| 61 | const size_t out_fifo_depth, // output fifo depth |
---|
[578] | 62 | |
---|
| 63 | const bool is_iob0, // cluster contains iob0 |
---|
| 64 | const bool is_iob1, // cluster contains iob1 |
---|
| 65 | const bool is_rsp, // only response router is modified |
---|
| 66 | const size_t l_width) // local field srcid width |
---|
[549] | 67 | : soclib::caba::BaseModule(name), |
---|
| 68 | |
---|
| 69 | p_clk( "p_clk" ), |
---|
| 70 | p_resetn( "p_resetn" ), |
---|
| 71 | p_in( alloc_elems<DspinInput<flit_width> >("p_in", 5) ), |
---|
| 72 | p_out( alloc_elems<DspinOutput<flit_width> >("p_out", 5) ), |
---|
| 73 | |
---|
| 74 | r_alloc_out( alloc_elems<sc_signal<bool> >("r_alloc_out", 5)), |
---|
| 75 | r_index_out( soclib::common::alloc_elems<sc_signal<size_t> >("r_index_out", 5)), |
---|
| 76 | r_fsm_in( alloc_elems<sc_signal<int> >("r_fsm_in", 5)), |
---|
| 77 | r_index_in( alloc_elems<sc_signal<size_t> >("r_index_in", 5)), |
---|
| 78 | |
---|
| 79 | m_local_x( x ), |
---|
| 80 | m_local_y( y ), |
---|
| 81 | |
---|
| 82 | m_x_width( x_width ), |
---|
| 83 | m_x_shift( flit_width - x_width ), |
---|
| 84 | m_x_mask( (0x1 << x_width) - 1 ), |
---|
| 85 | |
---|
| 86 | m_y_width( y_width ), |
---|
| 87 | m_y_shift( flit_width - x_width - y_width ), |
---|
| 88 | m_y_mask( (0x1 << y_width) - 1 ), |
---|
| 89 | |
---|
| 90 | m_l_width( l_width ), |
---|
| 91 | m_l_shift( flit_width - x_width - y_width - l_width ), |
---|
| 92 | m_l_mask( (0x1 << l_width) - 1 ), |
---|
| 93 | |
---|
[578] | 94 | m_is_iob0( is_iob0 ), |
---|
| 95 | m_is_iob1( is_iob1 ), |
---|
| 96 | m_is_rsp( is_rsp ) |
---|
[549] | 97 | |
---|
| 98 | { |
---|
| 99 | std::cout << " - Building DspinRouterTsar : " << name << std::endl; |
---|
| 100 | |
---|
| 101 | SC_METHOD (transition); |
---|
| 102 | dont_initialize(); |
---|
| 103 | sensitive << p_clk.pos(); |
---|
| 104 | |
---|
| 105 | SC_METHOD (genMoore); |
---|
| 106 | dont_initialize(); |
---|
| 107 | sensitive << p_clk.neg(); |
---|
| 108 | |
---|
| 109 | r_fifo_in = (GenericFifo<internal_flit_t>*) |
---|
| 110 | malloc(sizeof(GenericFifo<internal_flit_t>)*5); |
---|
| 111 | r_fifo_out = (GenericFifo<internal_flit_t>*) |
---|
| 112 | malloc(sizeof(GenericFifo<internal_flit_t>)*5); |
---|
| 113 | |
---|
| 114 | for( size_t i = 0 ; i < 5 ; i++ ) |
---|
| 115 | { |
---|
| 116 | std::ostringstream stri; |
---|
| 117 | stri << "r_in_fifo_" << i; |
---|
| 118 | new(&r_fifo_in[i]) |
---|
| 119 | GenericFifo<internal_flit_t >(stri.str(), in_fifo_depth); |
---|
| 120 | |
---|
| 121 | std::ostringstream stro; |
---|
| 122 | stro << "r_out_fifo_" << i; |
---|
| 123 | new(&r_fifo_out[i]) |
---|
| 124 | GenericFifo<internal_flit_t >(stro.str(), out_fifo_depth); |
---|
| 125 | } |
---|
| 126 | } // end constructor |
---|
| 127 | |
---|
| 128 | ////////////////////////////////////////////////// |
---|
| 129 | tmpl(size_t)::route( sc_uint<flit_width> data ) |
---|
| 130 | { |
---|
| 131 | size_t xdest = (size_t)(data >> m_x_shift) & m_x_mask; |
---|
| 132 | size_t ydest = (size_t)(data >> m_y_shift) & m_y_mask; |
---|
| 133 | size_t ldest = (size_t)(data >> m_l_shift) & m_l_mask; |
---|
| 134 | |
---|
| 135 | if (xdest < m_local_x ) return DSPIN_WEST; |
---|
| 136 | else if (xdest > m_local_x ) return DSPIN_EAST; |
---|
| 137 | else if (ydest < m_local_y ) return DSPIN_SOUTH; |
---|
| 138 | else if (ydest > m_local_y ) return DSPIN_NORTH; |
---|
| 139 | else // handling IOB0 & IOB1 special cases |
---|
| 140 | { |
---|
[578] | 141 | if (m_is_rsp and m_is_iob0 and (ldest > 0xA)) return DSPIN_WEST; |
---|
| 142 | else if (m_is_rsp and m_is_iob1 and (ldest > 0xA)) return DSPIN_EAST; |
---|
| 143 | else return DSPIN_LOCAL; |
---|
[549] | 144 | } |
---|
| 145 | } // end route() |
---|
| 146 | |
---|
| 147 | ///////////////////////// |
---|
| 148 | tmpl(void)::print_trace() |
---|
| 149 | { |
---|
| 150 | const char* port_name[] = {"NORTH","SOUTH","EAST ","WEST ","LOCAL"}; |
---|
| 151 | |
---|
| 152 | std::cout << "DSPIN_ROUTER " << name() << std::hex; |
---|
| 153 | for ( size_t out=0 ; out<5 ; out++) // loop on output ports |
---|
| 154 | { |
---|
| 155 | if ( r_alloc_out[out].read() ) |
---|
| 156 | { |
---|
| 157 | int in = r_index_out[out]; |
---|
| 158 | std::cout << " / " << port_name[in] << " -> " << port_name[out] ; |
---|
| 159 | } |
---|
| 160 | } |
---|
| 161 | std::cout << std::endl; |
---|
| 162 | } |
---|
| 163 | |
---|
| 164 | //////////////////////// |
---|
| 165 | tmpl(void)::transition() |
---|
| 166 | { |
---|
| 167 | // Long wires connecting input and output ports |
---|
| 168 | size_t req_in[5]; // input ports -> output ports |
---|
| 169 | size_t get_out[5]; // output ports -> input ports |
---|
| 170 | bool put_in[5]; // input ports -> output ports |
---|
| 171 | internal_flit_t flit_in[5]; // input ports -> output ports |
---|
| 172 | |
---|
| 173 | // control signals for the input fifos |
---|
| 174 | bool fifo_in_write[5]; |
---|
| 175 | bool fifo_in_read[5]; |
---|
| 176 | internal_flit_t fifo_in_wdata[5]; |
---|
| 177 | |
---|
| 178 | // control signals for the output fifos |
---|
| 179 | bool fifo_out_write[5]; |
---|
| 180 | bool fifo_out_read[5]; |
---|
| 181 | internal_flit_t fifo_out_wdata[5]; |
---|
| 182 | |
---|
| 183 | // Reset |
---|
| 184 | if ( p_resetn == false ) |
---|
| 185 | { |
---|
| 186 | for(size_t i = 0 ; i < 5 ; i++) |
---|
| 187 | { |
---|
| 188 | r_alloc_out[i] = false; |
---|
| 189 | r_index_out[i] = 0; |
---|
| 190 | r_index_in[i] = 0; |
---|
| 191 | r_fsm_in[i] = INFSM_IDLE; |
---|
| 192 | r_fifo_in[i].init(); |
---|
| 193 | r_fifo_out[i].init(); |
---|
| 194 | } |
---|
| 195 | return; |
---|
| 196 | } |
---|
| 197 | |
---|
| 198 | // fifos signals default values |
---|
| 199 | for(size_t i = 0 ; i < 5 ; i++) |
---|
| 200 | { |
---|
| 201 | fifo_in_read[i] = false; |
---|
| 202 | fifo_in_write[i] = p_in[i].write.read(); |
---|
| 203 | fifo_in_wdata[i].data = p_in[i].data.read(); |
---|
| 204 | fifo_in_wdata[i].eop = p_in[i].eop.read(); |
---|
| 205 | |
---|
| 206 | fifo_out_read[i] = p_out[i].read.read(); |
---|
| 207 | fifo_out_write[i] = false; |
---|
| 208 | } |
---|
| 209 | |
---|
| 210 | // loop on the output ports: |
---|
| 211 | // compute get_out[j] depending on the output port state |
---|
| 212 | // and combining fifo_out[j].wok and r_alloc_out[j] |
---|
| 213 | for ( size_t j = 0 ; j < 5 ; j++ ) |
---|
| 214 | { |
---|
| 215 | if( r_alloc_out[j].read() and (r_fifo_out[j].wok()) ) |
---|
| 216 | { |
---|
| 217 | get_out[j] = r_index_out[j].read(); |
---|
| 218 | } |
---|
| 219 | else |
---|
| 220 | { |
---|
| 221 | get_out[j] = 0xFFFFFFFF; |
---|
| 222 | } |
---|
| 223 | } |
---|
| 224 | |
---|
| 225 | // loop on the input ports : |
---|
| 226 | // The port state is defined by r_fsm_in[i], r_index_in[i] |
---|
| 227 | // The req_in[i] computation implements the X-FIRST algorithm. |
---|
| 228 | // Both put_in[i] and req_in[i] depend on the input port state. |
---|
| 229 | |
---|
| 230 | for ( size_t i = 0 ; i < 5 ; i++ ) |
---|
| 231 | { |
---|
| 232 | switch ( r_fsm_in[i].read() ) |
---|
| 233 | { |
---|
| 234 | case INFSM_IDLE: // no output port allocated |
---|
| 235 | { |
---|
| 236 | put_in[i] = false; |
---|
| 237 | if ( r_fifo_in[i].rok() ) // packet available in input fifo |
---|
| 238 | { |
---|
| 239 | req_in[i] = route( r_fifo_in[i].read().data ); |
---|
| 240 | r_index_in[i] = req_in[i]; |
---|
| 241 | r_fsm_in[i] = INFSM_REQ; |
---|
| 242 | } |
---|
| 243 | else |
---|
| 244 | { |
---|
| 245 | req_in[i] = 0xFFFFFFFF; // no request |
---|
| 246 | } |
---|
| 247 | break; |
---|
| 248 | } |
---|
| 249 | case INFSM_REQ: // waiting output port allocation |
---|
| 250 | { |
---|
| 251 | flit_in[i] = r_fifo_in[i].read(); |
---|
| 252 | put_in[i] = r_fifo_in[i].rok(); |
---|
| 253 | req_in[i] = r_index_in[i]; |
---|
| 254 | if ( get_out[r_index_in[i].read()] == i ) // first flit transfered |
---|
| 255 | { |
---|
| 256 | if ( r_fifo_in[i].read().eop ) r_fsm_in[i] = INFSM_IDLE; |
---|
| 257 | else r_fsm_in[i] = INFSM_ALLOC; |
---|
| 258 | } |
---|
| 259 | break; |
---|
| 260 | } |
---|
| 261 | case INFSM_ALLOC: // output port allocated |
---|
| 262 | { |
---|
| 263 | flit_in[i] = r_fifo_in[i].read(); |
---|
| 264 | put_in[i] = r_fifo_in[i].rok(); |
---|
| 265 | req_in[i] = 0xFFFFFFFF; // no request |
---|
| 266 | if ( r_fifo_in[i].read().eop and r_fifo_in[i].rok() and |
---|
| 267 | (get_out[r_index_in[i].read()] == i) ) // last flit transfered |
---|
| 268 | { |
---|
| 269 | r_fsm_in[i] = INFSM_IDLE; |
---|
| 270 | } |
---|
| 271 | break; |
---|
| 272 | } |
---|
| 273 | } // end switch |
---|
| 274 | } // end for input ports |
---|
| 275 | |
---|
| 276 | // loop on the output ports : |
---|
| 277 | // The r_alloc_out[j] and r_index_out[j] computation |
---|
| 278 | // implements the round-robin allocation policy. |
---|
| 279 | // These two registers implement a 10 states FSM. |
---|
| 280 | for( size_t j = 0 ; j < 5 ; j++ ) |
---|
| 281 | { |
---|
| 282 | if( not r_alloc_out[j].read() ) // not allocated: possible new allocation |
---|
| 283 | { |
---|
| 284 | for( size_t k = r_index_out[j].read() + 1 ; |
---|
| 285 | k < (r_index_out[j] + 6) ; k++) |
---|
| 286 | { |
---|
| 287 | size_t i = k % 5; |
---|
| 288 | |
---|
| 289 | if( req_in[i] == j ) |
---|
| 290 | { |
---|
| 291 | r_alloc_out[j] = true; |
---|
| 292 | r_index_out[j] = i; |
---|
| 293 | break; |
---|
| 294 | } |
---|
| 295 | } // end loop on input ports |
---|
| 296 | } |
---|
| 297 | else // allocated: possible desallocation |
---|
| 298 | { |
---|
| 299 | if ( flit_in[r_index_out[j]].eop and |
---|
| 300 | r_fifo_out[j].wok() and |
---|
| 301 | put_in[r_index_out[j]] ) |
---|
| 302 | { |
---|
| 303 | r_alloc_out[j] = false; |
---|
| 304 | } |
---|
| 305 | } |
---|
| 306 | } // end loop on output ports |
---|
| 307 | |
---|
| 308 | // loop on input ports : |
---|
| 309 | // fifo_in_read[i] computation (get data from fifo_in[i] |
---|
| 310 | // (computed here because it depends on get_out[]) |
---|
| 311 | for( size_t i = 0 ; i < 5 ; i++ ) |
---|
| 312 | { |
---|
| 313 | if ( r_fsm_in[i].read() != INFSM_IDLE ) |
---|
| 314 | { |
---|
| 315 | fifo_in_read[i] = (get_out[r_index_in[i].read()] == i); |
---|
| 316 | } |
---|
| 317 | else |
---|
| 318 | { |
---|
| 319 | fifo_in_read[i] = false; |
---|
| 320 | } |
---|
| 321 | } // end loop on input ports |
---|
| 322 | |
---|
| 323 | // loop on the output ports : |
---|
| 324 | // The fifo_out_write[j] and fifo_out_wdata[j] computation |
---|
| 325 | // implements the output port mux. |
---|
| 326 | for( size_t j = 0 ; j < 5 ; j++ ) |
---|
| 327 | { |
---|
| 328 | if( r_alloc_out[j] ) // output port allocated |
---|
| 329 | { |
---|
| 330 | fifo_out_write[j] = put_in[r_index_out[j]]; |
---|
| 331 | fifo_out_wdata[j] = flit_in[r_index_out[j]]; |
---|
| 332 | } |
---|
| 333 | } // end loop on the output ports |
---|
| 334 | |
---|
| 335 | // FIFOS update |
---|
| 336 | for(size_t i = 0 ; i < 5 ; i++) |
---|
| 337 | { |
---|
| 338 | r_fifo_in[i].update(fifo_in_read[i], |
---|
| 339 | fifo_in_write[i], |
---|
| 340 | fifo_in_wdata[i]); |
---|
| 341 | r_fifo_out[i].update(fifo_out_read[i], |
---|
| 342 | fifo_out_write[i], |
---|
| 343 | fifo_out_wdata[i]); |
---|
| 344 | } |
---|
| 345 | } // end transition |
---|
| 346 | |
---|
| 347 | //////////////////////////////// |
---|
| 348 | // genMoore |
---|
| 349 | //////////////////////////////// |
---|
| 350 | tmpl(void)::genMoore() |
---|
| 351 | { |
---|
| 352 | for(size_t i = 0 ; i < 5 ; i++) |
---|
| 353 | { |
---|
| 354 | // input ports : READ signals |
---|
| 355 | p_in[i].read = r_fifo_in[i].wok(); |
---|
| 356 | |
---|
| 357 | // output ports : DATA & WRITE signals |
---|
| 358 | p_out[i].data = r_fifo_out[i].read().data; |
---|
| 359 | p_out[i].eop = r_fifo_out[i].read().eop; |
---|
| 360 | p_out[i].write = r_fifo_out[i].rok(); |
---|
| 361 | } |
---|
| 362 | } // end genMoore |
---|
| 363 | |
---|
| 364 | }} // end namespace |
---|
| 365 | |
---|
| 366 | // Local Variables: |
---|
| 367 | // tab-width: 4 |
---|
| 368 | // c-basic-offset: 4 |
---|
| 369 | // c-file-offsets:((innamespace . 0)(inline-open . 0)) |
---|
| 370 | // indent-tabs-mode: nil |
---|
| 371 | // End: |
---|
| 372 | |
---|
| 373 | // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=4:softtabstop=4 |
---|