/////////////////////////////////////////////////////////////////////// // File: top.cpp // Author: Alain Greiner // Copyright: UPMC/LIP6 // Date : may 2013 // This program is released under the GNU public license ///////////////////////////////////////////////////////////////////////// // This file define a generic TSAR architecture. // The physical address space is 40 bits. // // The number of clusters cannot be larger than 256. // The number of processors per cluster cannot be larger than 8. // // - It uses four dspin_local_crossbar per cluster as local interconnect // - It uses two virtual_dspin routers per cluster as global interconnect // - It uses the vci_cc_vcache_wrapper // - It uses the vci_mem_cache // - It contains one vci_xicu per cluster. // - It contains one vci_multi_dma per cluster. // - It contains one vci_simple_ram per cluster to model the L3 cache. // // The communication between the MemCache and the Xram is 64 bits. // // All clusters are identical, but the cluster 0 (called io_cluster), // contains 6 extra components: // - the boot rom (BROM) // - the disk controller (BDEV) // - the multi-channel network controller (MNIC) // - the multi-channel chained buffer dma controller (CDMA) // - the multi-channel tty controller (MTTY) // - the frame buffer controller (FBUF) // // It is build with one single component implementing a cluster, // defined in files tsar_xbar_cluster.* (with * = cpp, h, sd) // // The IRQs are connected to XICUs as follow: // - The IRQ_IN[0] to IRQ_IN[7] ports are not used in all clusters. // - The DMA IRQs are connected to IRQ_IN[8] to IRQ_IN[15] in all clusters. // - The TTY IRQs are connected to IRQ_IN[16] to IRQ_IN[30] in I/O cluster. // - The BDEV IRQ is connected to IRQ_IN[31] in I/O cluster. // // Some hardware parameters are used when compiling the OS, and are used // by this top.cpp file. They must be defined in the hard_config.h file : // - CLUSTER_X : number of clusters in a row (power of 2) // - CLUSTER_Y : number of clusters in a column (power of 2) // - CLUSTER_SIZE : size of the segment allocated to a cluster // - NB_PROCS_MAX : number of processors per cluster (power of 2) // - NB_DMA_CHANNELS : number of DMA channels per cluster (< 9) // - NB_TTY_CHANNELS : number of TTY channels in I/O cluster (< 16) // - NB_NIC_CHANNELS : number of NIC channels in I/O cluster (< 9) // // Some other hardware parameters are not used when compiling the OS, // and can be directly defined in this top.cpp file: // - XRAM_LATENCY : external ram latency // - MEMC_WAYS : L2 cache number of ways // - MEMC_SETS : L2 cache number of sets // - L1_IWAYS // - L1_ISETS // - L1_DWAYS // - L1_DSETS // - FBUF_X_SIZE : width of frame buffer (pixels) // - FBUF_Y_SIZE : heigth of frame buffer (lines) // - BDEV_SECTOR_SIZE : block size for block drvice // - BDEV_IMAGE_NAME : file pathname for block device // - NIC_RX_NAME : file pathname for NIC received packets // - NIC_TX_NAME : file pathname for NIC transmited packets // - NIC_TIMEOUT : max number of cycles before closing a container ///////////////////////////////////////////////////////////////////////// // General policy for 40 bits physical address decoding: // All physical segments base addresses are multiple of 1 Mbytes // (=> the 24 LSB bits = 0, and the 16 MSB bits define the target) // The (x_width + y_width) MSB bits (left aligned) define // the cluster index, and the LADR bits define the local index: // | X_ID | Y_ID |---| LADR | OFFSET | // |x_width|y_width|---| 8 | 24 | ///////////////////////////////////////////////////////////////////////// // General policy for 14 bits SRCID decoding: // Each component is identified by (x_id, y_id, l_id) tuple. // | X_ID | Y_ID |---| L_ID | // |x_width|y_width|---| 6 | ///////////////////////////////////////////////////////////////////////// #include #include #include #include #include #include #include #include "gdbserver.h" #include "mapping_table.h" #include "alloc_elems.h" #include "tsar_super_cluster.h" //#define USE_ALMOS 1 #define USE_GIET #ifdef USE_ALMOS #ifdef USE_GIET #error "Can't use Two different OS" #endif #endif #ifndef USE_ALMOS #ifndef USE_GIET #error "You need to specify one OS" #endif #endif #ifdef USE_ALMOS #define PREFIX_OS "almos/" #include "almos/hard_config.h" #endif #ifdef USE_GIET #include "hard_config.h" #define RAM_TGTID 0 #define XCU_TGTID 1 #define DMA_TGTID 2 #define TTY_TGTID 3 #define IOC_TGTID 4 #define FBF_TGTID 5 #define NIC_TGTID 6 #define CMA_TGTID 7 #define ROM_TGTID 8 #define SIM_TGTID 9 #define FBF_X_SIZE FBUF_X_SIZE #define FBF_Y_SIZE FBUF_Y_SIZE #define Z_WIDTH3D 1 #define Z_SIZE3D (1 << Z_WIDTH3D) #define Z_IO3D (Y_IO & ((1 << Z_WIDTH3D) - 1)) #define Y_SIZE3D (Y_SIZE / Z_SIZE3D) #define Y_WIDTH3D (Y_WIDTH - Z_WIDTH3D) #define Y_IO3D (Y_IO >> Z_WIDTH3D) #define X_SIZE3D X_SIZE #define X_WIDTH3D X_WIDTH #define X_IO3D X_IO #define ELEVATOR_X 0 #define ELEVATOR_Y 0 #endif /////////////////////////////////////////////////// // Parallelisation /////////////////////////////////////////////////// #ifdef _OPENMP #include #endif // nluster index (computed from x,y coordinates) #ifdef USE_ALMOS #define cluster(x,y,z) (z + y * Z_SIZE3D + x * Y_SIZE3D) #else #define cluster(x,y,z) ((x << (Y_WIDTH3D + Z_WIDTH3D)) + (y << Z_WIDTH3D) + z) #endif #define min(x, y) (x < y ? x : y) /////////////////////////////////////////////////////////// // DSPIN parameters /////////////////////////////////////////////////////////// #define dspin_cmd_width 39 #define dspin_rsp_width 32 /////////////////////////////////////////////////////////// // VCI parameters /////////////////////////////////////////////////////////// #define vci_cell_width_int 4 #define vci_cell_width_ext 8 #ifdef USE_ALMOS #define vci_address_width 32 #endif #ifdef USE_GIET #define vci_address_width 40 #endif #define vci_plen_width 8 #define vci_rerror_width 1 #define vci_clen_width 1 #define vci_rflag_width 1 #define vci_srcid_width 14 #define vci_pktid_width 4 #define vci_trdid_width 4 #define vci_wrplen_width 1 //////////////////////////////////////////////////////////// // Secondary Hardware Parameters //////////////////////i///////////////////////////////////// #define XRAM_LATENCY 0 #define MEMC_WAYS 16 #define MEMC_SETS 256 #define L1_IWAYS 4 #define L1_ISETS 64 #define L1_DWAYS 4 #define L1_DSETS 64 #ifdef USE_ALMOS #define FBUF_X_SIZE 1024 #define FBUF_Y_SIZE 1024 #endif #ifdef USE_GIET #define BDEV_SECTOR_SIZE 512 #define BDEV_IMAGE_NAME "virt_hdd.dmg" #endif #ifdef USE_ALMOS #define BDEV_SECTOR_SIZE 4096 #define BDEV_IMAGE_NAME PREFIX_OS"hdd-img.bin" #endif #define NIC_RX_NAME PREFIX_OS"nic/rx_packets.txt" #define NIC_TX_NAME PREFIX_OS"nic/tx_packets.txt" #define NIC_TIMEOUT 10000 //////////////////////////////////////////////////////////// // Software to be loaded in ROM & RAM //////////////////////i///////////////////////////////////// #ifdef USE_ALMOS #define soft_name PREFIX_OS"preloader.elf" #endif #ifdef USE_GIET #define soft_name "../../softs/tsar_boot/preloader.elf" #endif //////////////////////////////////////////////////////////// // DEBUG Parameters default values //////////////////////i///////////////////////////////////// #define MAX_FROZEN_CYCLES 100000000 //////////////////////////////////////////////////////////////////// // TGTID definition in direct space // For all components: global TGTID = global SRCID = cluster_index //////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////// // Physical segments definition ///////////////////////////////////////////////////////// // There is 3 segments replicated in all clusters // and 5 specific segments in the "IO" cluster // (containing address 0xBF000000) ///////////////////////////////////////////////////////// #ifdef USE_ALMOS // 2^19 is the offset for the local id (8 bits for global ID : // 1 bit for Memcache or Peripheral, 4 for local peripheral id) // (Almos supports 32 bits physical addresses) #endif bool stop_called = false; ///////////////////////////////// int _main(int argc, char *argv[]) { using namespace sc_core; using namespace soclib::caba; using namespace soclib::common; const int64_t max_cycles = 5000000; // Maximum number of cycles simulated in one sc_start call int64_t ncycles = 0x7FFFFFFFFFFFFFFF; // simulated cycles char disk_name[256] = BDEV_IMAGE_NAME; // pathname to the disk image char nic_rx_name[256] = NIC_RX_NAME; // pathname to the rx packets file char nic_tx_name[256] = NIC_TX_NAME; // pathname to the tx packets file ssize_t threads_nr = 1; // simulator's threads number bool debug_ok = false; // trace activated size_t debug_period = 1; // trace period size_t debug_memc_id = 0; // index of memc to be traced size_t debug_proc_id = 0; // index of proc to be traced int64_t debug_from = 0; // trace start cycle int64_t frozen_cycles = MAX_FROZEN_CYCLES; // monitoring frozen processor int64_t reset_counters = -1; int64_t dump_counters = -1; bool do_reset_counters = false; bool do_dump_counters = false; struct timeval t1, t2; uint64_t ms1, ms2; ////////////// command line arguments ////////////////////// if (argc > 1) { for (int n = 1; n < argc; n = n + 2) { if ((strcmp(argv[n], "-NCYCLES") == 0) && (n + 1 < argc)) { ncycles = (int64_t) strtol(argv[n + 1], NULL, 0); } else if ((strcmp(argv[n],"-DISK") == 0) && (n + 1 < argc)) { strcpy(disk_name, argv[n + 1]); } else if ((strcmp(argv[n],"-DEBUG") == 0) && (n + 1 < argc)) { debug_ok = true; debug_from = (int64_t) strtol(argv[n + 1], NULL, 0); } else if ((strcmp(argv[n], "-MEMCID") == 0) && (n + 1 < argc)) { debug_memc_id = (size_t) strtol(argv[n + 1], NULL, 0); #ifdef USE_ALMOS assert((debug_memc_id < (X_SIZE3D * Y_SIZE3D)) && "debug_memc_id larger than X_SIZE3D * Y_SIZE3D" ); #else size_t x = debug_memc_id >> (Y_WIDTH3D + Z_WIDTH3D); size_t y = (debug_memc_id >> Z_WIDTH3D) & ((1 << Y_WIDTH3D) - 1); size_t z = debug_memc_id & ((1 << Z_WIDTH3D) - 1); assert( (x <= X_SIZE3D) and (y <= Y_SIZE3D) and (z <= Z_SIZE3D) && "MEMCID parameter refers a not valid memory cache"); #endif } else if ((strcmp(argv[n], "-PROCID") == 0) && (n + 1 < argc)) { debug_proc_id = (size_t) strtol(argv[n + 1], NULL, 0); #ifdef USE_ALMOS assert((debug_proc_id < (X_SIZE3D * Y_SIZE3D * Z_SIZE3D * NB_PROCS_MAX)) && "debug_proc_id larger than X_SIZE3D * Y_SIZE3D * Z_SIZE3D * NB_PROCS"); #else size_t cluster_xyz = debug_proc_id / NB_PROCS_MAX ; size_t x = cluster_xyz >> (Y_WIDTH3D + Z_WIDTH3D); size_t y = (cluster_xyz >> Z_WIDTH3D) & ((1 << Y_WIDTH3D) - 1); size_t z = cluster_xyz & ((1 << Z_WIDTH3D) - 1); assert( (x <= X_SIZE3D) and (y <= Y_SIZE3D) and (z <= Z_SIZE3D) && "PROCID parameter refers a not valid processor"); #endif } else if ((strcmp(argv[n], "-THREADS") == 0) && ((n + 1) < argc)) { threads_nr = (ssize_t) strtol(argv[n + 1], NULL, 0); threads_nr = (threads_nr < 1) ? 1 : threads_nr; } else if ((strcmp(argv[n], "-FROZEN") == 0) && (n + 1 < argc)) { frozen_cycles = (int64_t) strtol(argv[n + 1], NULL, 0); } else if ((strcmp(argv[n], "-PERIOD") == 0) && (n + 1 < argc)) { debug_period = (size_t) strtol(argv[n + 1], NULL, 0); } else if ((strcmp(argv[n], "--reset-counters") == 0) && (n + 1 < argc)) { reset_counters = (int64_t) strtol(argv[n + 1], NULL, 0); do_reset_counters = true; } else if ((strcmp(argv[n], "--dump-counters") == 0) && (n + 1 < argc)) { dump_counters = (int64_t) strtol(argv[n + 1], NULL, 0); do_dump_counters = true; } else { std::cout << " Arguments are (key,value) couples." << std::endl; std::cout << " The order is not important." << std::endl; std::cout << " Accepted arguments are :" << std::endl << std::endl; std::cout << " -SOFT pathname_for_embedded_soft" << std::endl; std::cout << " -DISK pathname_for_disk_image" << std::endl; std::cout << " -NCYCLES number_of_simulated_cycles" << std::endl; std::cout << " -DEBUG debug_start_cycle" << std::endl; std::cout << " -THREADS simulator's threads number" << std::endl; std::cout << " -FROZEN max_number_of_lines" << std::endl; std::cout << " -PERIOD number_of_cycles between trace" << std::endl; std::cout << " -MEMCID index_memc_to_be_traced" << std::endl; std::cout << " -PROCID index_proc_to_be_traced" << std::endl; exit(0); } } } // checking hardware parameters assert( ( (X_SIZE3D == 1) or (X_SIZE3D == 2) or (X_SIZE3D == 4) or (X_SIZE3D == 8) or (X_SIZE3D == 16) ) and "The X_SIZE3D parameter must be 1, 2, 4, 8 or 16" ); assert( ( (Y_SIZE3D == 1) or (Y_SIZE3D == 2) or (Y_SIZE3D == 4) or (Y_SIZE3D == 8) or (Y_SIZE3D == 16) ) and "The Y_SIZE3D parameter must be 1, 2, 4, 8 or 16" ); assert( ( (Z_SIZE3D == 1) or (Z_SIZE3D == 2) or (Z_SIZE3D == 4) or (Z_SIZE3D == 8) or (Z_SIZE3D == 16) ) and "The Z_SIZE3D parameter must be 1, 2, 4, 8 or 16" ); assert( ( (NB_PROCS_MAX == 1) or (NB_PROCS_MAX == 2) or (NB_PROCS_MAX == 4) or (NB_PROCS_MAX == 8) ) and "The NB_PROCS_MAX parameter must be 1, 2, 4 or 8" ); assert( (NB_DMA_CHANNELS < 9) and "The NB_DMA_CHANNELS parameter must be smaller than 9" ); assert( (NB_TTY_CHANNELS < 15) and "The NB_TTY_CHANNELS parameter must be smaller than 15" ); assert( (NB_NIC_CHANNELS < 9) and "The NB_NIC_CHANNELS parameter must be smaller than 9" ); #ifdef USE_GIET assert( (vci_address_width == 40) and "VCI address width with the GIET must be 40 bits" ); #endif #ifdef USE_ALMOS assert( (vci_address_width == 32) and "VCI address width with ALMOS must be 32 bits" ); #endif std::cout << std::endl; std::cout << " - X_SIZE3D = " << X_SIZE3D << std::endl; std::cout << " - Y_SIZE3D = " << Y_SIZE3D << std::endl; std::cout << " - Z_SIZE3D = " << Z_SIZE3D << std::endl; std::cout << " - NB_PROCS_MAX = " << NB_PROCS_MAX << std::endl; std::cout << " - NB_DMA_CHANNELS = " << NB_DMA_CHANNELS << std::endl; std::cout << " - NB_TTY_CHANNELS = " << NB_TTY_CHANNELS << std::endl; std::cout << " - NB_NIC_CHANNELS = " << NB_NIC_CHANNELS << std::endl; std::cout << " - MEMC_WAYS = " << MEMC_WAYS << std::endl; std::cout << " - MEMC_SETS = " << MEMC_SETS << std::endl; std::cout << " - RAM_LATENCY = " << XRAM_LATENCY << std::endl; std::cout << " - MAX_FROZEN = " << frozen_cycles << std::endl; std::cout << std::endl; // Internal and External VCI parameters definition typedef soclib::caba::VciParams vci_param_int; typedef soclib::caba::VciParams vci_param_ext; #ifdef _OPENMP omp_set_dynamic(false); omp_set_num_threads(threads_nr); std::cerr << "Built with openmp version " << _OPENMP << std::endl; #endif // Define parameters depending on mesh size size_t x_width; size_t y_width; size_t z_width; #ifdef USE_ALMOS if (X_SIZE3D == 1) x_width = 0; else if (X_SIZE3D == 2) x_width = 1; else if (X_SIZE3D <= 4) x_width = 2; else if (X_SIZE3D <= 8) x_width = 3; else x_width = 4; if (Y_SIZE3D == 1) y_width = 0; else if (Y_SIZE3D == 2) y_width = 1; else if (Y_SIZE3D <= 4) y_width = 2; else if (Y_SIZE3D <= 8) y_width = 3; else y_width = 4; if (Z_SIZE3D == 1) z_width = 0; else if (Z_SIZE3D == 2) z_width = 1; else if (Z_SIZE3D <= 4) z_width = 2; else if (Z_SIZE3D <= 8) z_width = 3; else z_width = 4; #else x_width = X_WIDTH3D; y_width = Y_WIDTH3D; z_width = Z_WIDTH3D; assert(((X_WIDTH3D + Y_WIDTH3D + Z_WIDTH3D) <= 8) and "Up to 256 clusters"); assert((X_SIZE3D <= (1 << X_WIDTH3D)) and (Y_SIZE3D <= (1 << Y_WIDTH3D)) and (Z_SIZE3D <= (1 << Z_WIDTH3D)) and "The X_WIDTH3D and Y_WIDTH3D and Z_WIDTH3D parameter are insufficient"); #endif ///////////////////// // Mapping Tables ///////////////////// // internal network MappingTable maptabd(vci_address_width, IntTab(x_width + y_width + z_width, 16 - x_width - y_width - z_width), IntTab(x_width + y_width + z_width, vci_srcid_width - x_width - y_width - z_width), 0x00FF000000); for (size_t x = 0; x < X_SIZE3D; x++) { for (size_t y = 0; y < Y_SIZE3D; y++) { for (size_t z = 0; z < Z_SIZE3D; z++) { sc_uint offset; offset = (sc_uint) cluster(x,y,z) << (vci_address_width - x_width - y_width - z_width); std::ostringstream si; si << "seg_xicu_" << x << "_" << y << "_" << z; maptabd.add(Segment(si.str(), SEG_XCU_BASE + offset, SEG_XCU_SIZE, IntTab(cluster(x,y,z), XCU_TGTID), false)); std::ostringstream sd; sd << "seg_mdma_" << x << "_" << y << "_" << z; maptabd.add(Segment(sd.str(), SEG_DMA_BASE + offset, SEG_DMA_SIZE, IntTab(cluster(x,y,z), DMA_TGTID), false)); std::ostringstream sh; sh << "seg_memc_" << x << "_" << y << "_" << z; maptabd.add(Segment(sh.str(), SEG_RAM_BASE + offset, SEG_RAM_SIZE, IntTab(cluster(x,y,z), RAM_TGTID), true)); if (x == X_IO3D && y == Y_IO3D && z == Z_IO3D) { maptabd.add(Segment("seg_mtty", SEG_TTY_BASE, SEG_TTY_SIZE, IntTab(cluster(x,y,z),TTY_TGTID), false)); maptabd.add(Segment("seg_fbuf", SEG_FBF_BASE, SEG_FBF_SIZE, IntTab(cluster(x,y,z),FBF_TGTID), false)); maptabd.add(Segment("seg_bdev", SEG_IOC_BASE, SEG_IOC_SIZE, IntTab(cluster(x,y,z),IOC_TGTID), false)); maptabd.add(Segment("seg_brom", SEG_ROM_BASE, SEG_ROM_SIZE, IntTab(cluster(x,y,z),ROM_TGTID), true)); maptabd.add(Segment("seg_mnic", SEG_NIC_BASE, SEG_NIC_SIZE, IntTab(cluster(x,y,z),NIC_TGTID), false)); maptabd.add(Segment("seg_cdma", SEG_CMA_BASE, SEG_CMA_SIZE, IntTab(cluster(x,y,z),CMA_TGTID), false)); maptabd.add(Segment("seg_simh", SEG_SIM_BASE, SEG_SIM_SIZE, IntTab(cluster(x,y,z),SIM_TGTID), false)); } } } } std::cout << maptabd << std::endl; // external network MappingTable maptabx(vci_address_width, IntTab(x_width + y_width + z_width), IntTab(x_width + y_width + z_width), 0xFFFF000000ULL); for (size_t x = 0; x < X_SIZE3D; x++) { for (size_t y = 0; y < Y_SIZE3D ; y++) { for (size_t z = 0; z < Z_SIZE3D ; z++) { sc_uint offset; offset = (sc_uint) cluster(x,y,z) << (vci_address_width - x_width - y_width - z_width); std::ostringstream sh; sh << "x_seg_memc_" << x << "_" << y << "_" << z; maptabx.add(Segment(sh.str(), SEG_RAM_BASE + offset, SEG_RAM_SIZE, IntTab(cluster(x,y,z)), false)); } } } std::cout << maptabx << std::endl; //////////////////// // Signals /////////////////// sc_clock signal_clk("clk"); sc_signal signal_resetn("resetn"); // Z-axis inter-clusters DSPIN signals DspinSignals* signal_dspin_z_cmd_inc = alloc_elems >("signal_dspin_h_cmd_inc", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_cmd_dec = alloc_elems >("signal_dspin_h_cmd_dec", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_rsp_inc = alloc_elems >("signal_dspin_h_rsp_inc", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_rsp_dec = alloc_elems >("signal_dspin_h_rsp_dec", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_m2p_inc = alloc_elems >("signal_dspin_h_m2p_inc", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_m2p_dec = alloc_elems >("signal_dspin_h_m2p_dec", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_p2m_inc = alloc_elems >("signal_dspin_h_p2m_inc", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_p2m_dec = alloc_elems >("signal_dspin_h_p2m_dec", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_cla_inc = alloc_elems >("signal_dspin_h_cla_inc", Z_SIZE3D + 1); DspinSignals* signal_dspin_z_cla_dec = alloc_elems >("signal_dspin_h_cla_dec", Z_SIZE3D + 1); //////////////////////////// // Loader //////////////////////////// soclib::common::Loader loader(soft_name); typedef soclib::common::GdbServer proc_iss; proc_iss::set_loader(loader); //////////////////////////// // Clusters construction //////////////////////////// TsarSuperCluster * clusters[Z_SIZE3D]; for (size_t z = 0; z < Z_SIZE3D; z++) { std::cout << std::endl; std::cout << "SuperCluster_" << z << std::endl; std::ostringstream sc; sc << "scluster_" << z; clusters[z] = new TsarSuperCluster ( sc.str().c_str(), NB_PROCS_MAX, NB_TTY_CHANNELS, NB_DMA_CHANNELS, X_SIZE3D, Y_SIZE3D, z, ELEVATOR_X, ELEVATOR_Y, maptabd, maptabx, x_width, y_width, z_width, P_WIDTH, vci_srcid_width, RAM_TGTID, XCU_TGTID, DMA_TGTID, FBF_TGTID, TTY_TGTID, ROM_TGTID, NIC_TGTID, CMA_TGTID, IOC_TGTID, SIM_TGTID, MEMC_WAYS, MEMC_SETS, L1_IWAYS, L1_ISETS, L1_DWAYS, L1_DSETS, IRQ_PER_PROCESSOR, XRAM_LATENCY, X_IO3D, Y_IO3D, Z_IO3D, FBF_X_SIZE, FBF_Y_SIZE, disk_name, BDEV_SECTOR_SIZE, NB_NIC_CHANNELS, nic_rx_name, nic_tx_name, NIC_TIMEOUT, NB_CMA_CHANNELS, loader, frozen_cycles, debug_from, debug_ok, debug_ok ); } /////////////////////////////////////////////////////////////// // Net-list /////////////////////////////////////////////////////////////// for (int z = 0; z < Z_SIZE3D; z++) { // Clock & RESET clusters[z]->p_clk (signal_clk); clusters[z]->p_resetn (signal_resetn); // Inter Clusters Z connections #define UP 0 #define DOWN 1 clusters[z]->p_cmd_out[UP] (signal_dspin_z_cmd_inc[z + 1]); clusters[z]->p_cmd_in[DOWN] (signal_dspin_z_cmd_inc[z]); clusters[z]->p_cmd_in[UP] (signal_dspin_z_cmd_dec[z + 1]); clusters[z]->p_cmd_out[DOWN] (signal_dspin_z_cmd_dec[z]); clusters[z]->p_rsp_out[UP] (signal_dspin_z_rsp_inc[z + 1]); clusters[z]->p_rsp_in[DOWN] (signal_dspin_z_rsp_inc[z]); clusters[z]->p_rsp_in[UP] (signal_dspin_z_rsp_dec[z + 1]); clusters[z]->p_rsp_out[DOWN] (signal_dspin_z_rsp_dec[z]); clusters[z]->p_m2p_out[UP] (signal_dspin_z_m2p_inc[z + 1]); clusters[z]->p_m2p_in[DOWN] (signal_dspin_z_m2p_inc[z]); clusters[z]->p_m2p_in[UP] (signal_dspin_z_m2p_dec[z + 1]); clusters[z]->p_m2p_out[DOWN] (signal_dspin_z_m2p_dec[z]); clusters[z]->p_p2m_out[UP] (signal_dspin_z_p2m_inc[z + 1]); clusters[z]->p_p2m_in[DOWN] (signal_dspin_z_p2m_inc[z]); clusters[z]->p_p2m_in[UP] (signal_dspin_z_p2m_dec[z + 1]); clusters[z]->p_p2m_out[DOWN] (signal_dspin_z_p2m_dec[z]); clusters[z]->p_cla_out[UP] (signal_dspin_z_cla_inc[z + 1]); clusters[z]->p_cla_in[DOWN] (signal_dspin_z_cla_inc[z]); clusters[z]->p_cla_in[UP] (signal_dspin_z_cla_dec[z + 1]); clusters[z]->p_cla_out[DOWN] (signal_dspin_z_cla_dec[z]); } std::cout << std::endl << "Z connections done" << std::endl; #ifdef WT_IDL std::list > * > l1_caches; for (int x = 0; x < X_SIZE3D; x++) { for (int y = 0; y < Y_SIZE3D; y++) { for (int proc = 0; proc < NB_PROCS_MAX; proc++) { l1_caches.push_back(clusters[x][y]->proc[proc]); } } } for (int x = 0; x < X_SIZE3D; x++) { for (int y = 0; y < Y_SIZE3D; y++) { clusters[x][y]->memc->set_vcache_list(l1_caches); } } #endif // #define SC_TRACE #ifdef SC_TRACE sc_trace_file * tf = sc_create_vcd_trace_file("my_trace_file"); #if 0 for (int x = 0; x < X_SIZE3D - 1; x++) { for (int y = 0; y < Y_SIZE3D; y++) { for (int k = 0; k < 3; k++) { signal_dspin_h_cmd_inc[x][y][k].trace(tf, "dspin_h_cmd_inc"); signal_dspin_h_cmd_dec[x][y][k].trace(tf, "dspin_h_cmd_dec"); } for (int k = 0; k < 2; k++) { signal_dspin_h_rsp_inc[x][y][k].trace(tf, "dspin_h_rsp_inc"); signal_dspin_h_rsp_dec[x][y][k].trace(tf, "dspin_h_rsp_dec"); } } } for (int y = 0; y < Y_SIZE3D - 1; y++) { for (int x = 0; x < X_SIZE3D; x++) { for (int k = 0; k < 3; k++) { signal_dspin_v_cmd_inc[x][y][k].trace(tf, "dspin_v_cmd_inc"); signal_dspin_v_cmd_dec[x][y][k].trace(tf, "dspin_v_cmd_dec"); } for (int k = 0; k < 2; k++) { signal_dspin_v_rsp_inc[x][y][k].trace(tf, "dspin_v_rsp_inc"); signal_dspin_v_rsp_dec[x][y][k].trace(tf, "dspin_v_rsp_dec"); } } } for (int x = 0; x < (X_SIZE3D); x++) { for (int y = 0; y < Y_SIZE3D; y++) { std::ostringstream signame; signame << "cluster" << x << "_" << y; clusters[x][y]->trace(tf, signame.str()); } } #endif for (size_t z = 0; z < (Z_SIZE3D); z++) { clusters[z]->trace(tf); } #endif //////////////////////////////////////////////////////// // Simulation /////////////////////////////////////////////////////// sc_start(sc_core::sc_time(0, SC_NS)); signal_resetn = false; // set network boundaries signals default values // for all boundary clusters signal_dspin_z_cmd_inc[0].write = false; signal_dspin_z_cmd_inc[0].read = true; signal_dspin_z_cmd_dec[0].write = false; signal_dspin_z_cmd_dec[0].read = true; signal_dspin_z_cmd_inc[Z_SIZE3D].write = false; signal_dspin_z_cmd_inc[Z_SIZE3D].read = true; signal_dspin_z_cmd_dec[Z_SIZE3D].write = false; signal_dspin_z_cmd_dec[Z_SIZE3D].read = true; signal_dspin_z_rsp_inc[0].write = false; signal_dspin_z_rsp_inc[0].read = true; signal_dspin_z_rsp_dec[0].write = false; signal_dspin_z_rsp_dec[0].read = true; signal_dspin_z_rsp_inc[Z_SIZE3D].write = false; signal_dspin_z_rsp_inc[Z_SIZE3D].read = true; signal_dspin_z_rsp_dec[Z_SIZE3D].write = false; signal_dspin_z_rsp_dec[Z_SIZE3D].read = true; signal_dspin_z_p2m_inc[0].write = false; signal_dspin_z_p2m_inc[0].read = true; signal_dspin_z_p2m_dec[0].write = false; signal_dspin_z_p2m_dec[0].read = true; signal_dspin_z_p2m_inc[Z_SIZE3D].write = false; signal_dspin_z_p2m_inc[Z_SIZE3D].read = true; signal_dspin_z_p2m_dec[Z_SIZE3D].write = false; signal_dspin_z_p2m_dec[Z_SIZE3D].read = true; signal_dspin_z_m2p_inc[0].write = false; signal_dspin_z_m2p_inc[0].read = true; signal_dspin_z_m2p_dec[0].write = false; signal_dspin_z_m2p_dec[0].read = true; signal_dspin_z_m2p_inc[Z_SIZE3D].write = false; signal_dspin_z_m2p_inc[Z_SIZE3D].read = true; signal_dspin_z_m2p_dec[Z_SIZE3D].write = false; signal_dspin_z_m2p_dec[Z_SIZE3D].read = true; signal_dspin_z_cla_inc[0].write = false; signal_dspin_z_cla_inc[0].read = true; signal_dspin_z_cla_dec[0].write = false; signal_dspin_z_cla_dec[0].read = true; signal_dspin_z_cla_inc[Z_SIZE3D].write = false; signal_dspin_z_cla_inc[Z_SIZE3D].read = true; signal_dspin_z_cla_dec[Z_SIZE3D].write = false; signal_dspin_z_cla_dec[Z_SIZE3D].read = true; for (int z = 0; z < Z_SIZE3D; z++) { clusters[z]->reset(); } sc_start(sc_core::sc_time(1, SC_NS)); signal_resetn = true; if (debug_ok) { if (gettimeofday(&t1, NULL) != 0) { perror("gettimeofday"); return EXIT_FAILURE; } for (int64_t n = 1; n < ncycles && !stop_called; n++) { if ((n % max_cycles) == 0) { if (gettimeofday(&t2, NULL) != 0) { perror("gettimeofday"); return EXIT_FAILURE; } ms1 = (uint64_t) t1.tv_sec * 1000ULL + (uint64_t) t1.tv_usec / 1000; ms2 = (uint64_t) t2.tv_sec * 1000ULL + (uint64_t) t2.tv_usec / 1000; std::cerr << "platform clock frequency " << (double) 5000000 / (double) (ms2 - ms1) << "Khz" << std::endl; if (gettimeofday(&t1, NULL) != 0) { perror("gettimeofday"); return EXIT_FAILURE; } } if ((n > debug_from) and (n % debug_period == 0)) { std::cout << "****************** cycle " << std::dec << n ; std::cout << "************************************************" << std::endl; } sc_start(sc_core::sc_time(1, SC_NS)); } } else { int64_t n = 0; while (!stop_called && n != ncycles) { if (gettimeofday(&t1, NULL) != 0) { perror("gettimeofday"); return EXIT_FAILURE; } int64_t nb_cycles = min(max_cycles, ncycles - n); if (do_reset_counters) { nb_cycles = min(nb_cycles, reset_counters - n); } if (do_dump_counters) { nb_cycles = min(nb_cycles, dump_counters - n); } sc_start(sc_core::sc_time(nb_cycles, SC_NS)); n += nb_cycles; if (gettimeofday(&t2, NULL) != 0) { perror("gettimeofday"); return EXIT_FAILURE; } ms1 = (uint64_t) t1.tv_sec * 1000ULL + (uint64_t) t1.tv_usec / 1000; ms2 = (uint64_t) t2.tv_sec * 1000ULL + (uint64_t) t2.tv_usec / 1000; std::cerr << std::dec << "cycle " << n << " platform clock frequency " << (double) nb_cycles / (double) (ms2 - ms1) << "Khz" << std::endl; } } return EXIT_SUCCESS; } void handler(int dummy = 0) { stop_called = true; sc_stop(); } void voidhandler(int dummy = 0) {} int sc_main (int argc, char *argv[]) { signal(SIGINT, handler); //signal(SIGPIPE, voidhandler); try { return _main(argc, argv); } catch (std::exception &e) { std::cout << e.what() << std::endl; } catch (...) { std::cout << "Unknown exception occured" << std::endl; throw; } return 1; } // Local Variables: // tab-width: 3 // c-basic-offset: 3 // c-file-offsets:((innamespace . 0)(inline-open . 0)) // indent-tabs-mode: nil // End: // vim: filetype=cpp:expandtab:shiftwidth=3:tabstop=3:softtabstop=3