--- /dev/null
+package dol.visitor.cell;
+
+import java.io.FileOutputStream;
+import java.io.OutputStream;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.Vector;
+
+import dol.datamodel.pn.Channel;
+import dol.datamodel.pn.Port;
+import dol.datamodel.pn.Process;
+import dol.datamodel.pn.ProcessNetwork;
+import dol.visitor.PNVisitor;
+import dol.main.UserInterface;
+import dol.util.CodePrintStream;
+
+/**
+ * This class is a class for a visitor that is used to generate the main
+ * program.
+ *
+ * @author lschor, 2008-10-30
+ *
+ * Remarks: Based on a original file from khuang for rtems
+ *
+ * Revision: 2008-10-30: Updated the file for the CBE 2008-11-08: Add
+ * double buffering 2008-11-16: Add new fifo implementation and defines
+ * for measurement 2008-11-21: Sink/Source do not run on the SPE, but on
+ * the PPE (as Linux thread) 2009-03-17: Added Protothread Support
+ * 2009-04-07: Several modifications, so that one only writes if there
+ * is enough memory, do not allocate all queues on PPU
+ */
+public class CellModuleVisitor extends PNVisitor {
+
+ /**
+ * Constructor.
+ *
+ * @param dir
+ * path of this file
+ */
+ public CellModuleVisitor(String dir, HashMap<Port, Integer> portMap,
+ CellMapping mapping) {
+ _dir = dir;
+ _portMap = portMap;
+ _mapping = mapping;
+ }
+
+ /**
+ * Visit process network.
+ *
+ * @param x
+ * process network that needs to be rendered
+ */
+ public void visitComponent(ProcessNetwork x) {
+ try {
+ // Create in a first step the different SPU OS Layers
+ ArrayList<Vector<Process>> spuList = _mapping.getSPUList();
+ Vector<Process> ppu = _mapping.getPPU();
+
+ // Create the main file for the PPU
+ _ui = UserInterface.getInstance();
+ String filename = _dir + _delimiter + "ppu_main.cpp";
+ OutputStream file = new FileOutputStream(filename);
+ _mainPS = new CodePrintStream(file);
+
+ // create header section
+ _mainPS.println("// ========================");
+ _mainPS.println("// ppu_main.cpp file");
+ _mainPS.println("// ========================");
+ _mainPS.println("");
+
+ // Includes
+ _mainPS.println("#include \"ppu_main.h\"");
+ _mainPS.println("");
+
+ // Function to create and run one SPE thread
+ _mainPS.println("// create and run one SPE thread");
+ _mainPS.println("void *spu_pthread(void *arg) {");
+ _mainPS.println(" spu_data_t *datp = (spu_data_t *)arg;");
+ _mainPS.println(" uint32_t entry = SPE_DEFAULT_ENTRY;");
+ _mainPS.println(" printf(\")PPE: spe thread start run\\n\" );");
+ _mainPS.println(" if(spe_context_run(datp->spe_ctx,&entry,0,datp->argp,NULL,NULL)<0){");
+ _mainPS.println(" perror (\"Failed running context\"); exit (1);");
+ _mainPS.println(" }");
+ _mainPS.println(" printf(\")PPE: spe thread finish run\\n\");");
+ _mainPS.println(" pthread_exit(NULL);");
+ _mainPS.println("}");
+ _mainPS.println("");
+
+ // Declaration of the Header function for the PPE-Wrappers
+ /*
+ * for (Process p : ppu) { _mainPS.println("void *"+ p.getBasename()
+ * + "_wrapper( void *ptr );"); }
+ */
+ _mainPS.println("void *ppu( void *ptr );");
+ _mainPS.println();
+
+ // Create the port_id and the port_queue_id arrays to send over the
+ // DMA
+ for (Process process : _mapping.getAllSpuProcess()) {
+ String processName = process.getName();
+ _mainPS.println("volatile char " + processName
+ + "_name[256] __attribute__ ((aligned(16)));");
+ }
+ _mainPS.println();
+
+ // Go through all SPEs
+ for (Vector<Process> spu : spuList) {
+ _mainPS.println("volatile uint64_t spu_"
+ + spuList.indexOf(spu) + "[" + spu.size()
+ + "] __attribute__ ((aligned(16)));");
+ }
+ _mainPS.println();
+ _mainPS.println("// The input queue of a process is always on the SPU of the receiver, negative means that the queue is on the PPE");
+ _mainPS.println("volatile int32_t queueOnSPU [NUM_FIFO] __attribute__ ((aligned(16)));");
+ _mainPS.println();
+ _mainPS.println(" // The queue comes from this SPU");
+ _mainPS.println("volatile int32_t queueFromSPU [NUM_FIFO] __attribute__ ((aligned(16)));");
+ _mainPS.println();
+ _mainPS.println(" volatile uint64_t ea_ls_base[NUM_SPES] __attribute__ ((aligned(16)));");
+ _mainPS.println("volatile uint64_t context_addr[NUM_SPES] __attribute__ ((aligned(16)));");
+ _mainPS.println("volatile spe_spu_control_area_t* mfc_ctl[NUM_SPES] __attribute__ ((aligned(16)));");
+ _mainPS.println("volatile uint64_t fifoTails[NUM_FIFO] __attribute__ ((aligned(16)));");
+ _mainPS.println();
+
+ // Create the main function
+ _mainPS.println("int main()");
+ _mainPS.println("{");
+
+ // List with all SPU to be open
+ _mainPS.println(" char spe_names[NUM_SPES][60] = {");
+ for (Vector<Process> spu : spuList) {
+ _mainPS.println(" \"spu/spu_os_" + spuList.indexOf(spu)
+ + "\",");
+ }
+ _mainPS.println(" };");
+
+ _mainPS.println();
+
+ for (Channel c : x.getChannelList()) {
+ // Search for the origin spe
+ int origin = -1;
+
+ for (Vector<Process> spu : spuList) {
+
+ if (spu.contains(c.getOrigin())) {
+ origin = spuList.indexOf(spu);
+ break;
+ }
+ }
+
+ // Origin on the PPU
+ if (origin == -1) {
+ origin = (-1) * (ppu.indexOf(c.getOrigin()) + 1);
+ }
+
+ _mainPS.println(" queueFromSPU["
+ + x.getChannelList().indexOf(c) + "] = " + origin
+ + ";");
+ }
+
+ _mainPS.println();
+
+ for (Channel c : x.getChannelList()) {
+ // Search for the origin spe
+ int origin = -1;
+
+ for (Vector<Process> spu : spuList) {
+ if (spu.contains(c.getTarget())) {
+ origin = spuList.indexOf(spu);
+ break;
+ }
+ }
+
+ // Origin on the PPU
+ if (origin == -1) {
+ origin = (-1) * (ppu.indexOf(c.getTarget()) + 1);
+ }
+
+ _mainPS.println(" queueOnSPU["
+ + x.getChannelList().indexOf(c) + "] = " + origin
+ + ";");
+ }
+
+ _mainPS.println();
+
+ // connect ports to channels
+ HashMap<Channel, Integer> channel_map = new HashMap<Channel, Integer>();
+
+ int j = 0;
+ for (Channel c : x.getChannelList()) {
+ channel_map.put(c, j++);
+ }
+
+ // Add to each process the ports and the queues
+ _mainPS
+ .println(" //Add to each process the ports and the queues");
+ j = 0;
+
+ // SPU Process
+ for (Process process : _mapping.getAllSpuProcess()) {
+ String processName = process.getName();
+ _mainPS.println(" ctx_proc[" + j + "]"
+ + ".is_detached = 0;");
+ _mainPS.println(" strcpy((char *)" + processName
+ + "_name, " + "\"" + processName + "\");");
+ _mainPS.println(" ctx_proc[" + j + "]"
+ + ".processName = (uint64_t) " + processName
+ + "_name;");
+ _mainPS.println(" ctx_proc[" + j + "]"
+ + ".processNameLen = ((strlen((char *)"
+ + processName + "_name) + 15) & ~15);");
+ _mainPS.println();
+ j++;
+ }
+
+ _mainPS.println(" // Mapping tasks -> What to map to which SPE?");
+ _mainPS.println(" // Example: Square 0-2 to SPE 0 AND Square 3-5 to SPE 1");
+ j = 0;
+ for (Process process : _mapping.getAllSpuProcess()) {
+ for (Vector<Process> spu : spuList) {
+ if (spu.contains(process)) {
+ _mainPS.println(" spu_" + spuList.indexOf(spu)
+ + "[" + spu.indexOf(process)
+ + "] = (uint64_t) &(ctx_proc[" + j
+ + "]);");
+ j++;
+ break;
+ }
+ }
+ }
+
+ for (Vector<Process> spu : spuList) {
+ _mainPS.println(" ctx_spu[" + spuList.indexOf(spu)
+ + "].procContents = (uint64_t) spu_"
+ + spuList.indexOf(spu) + ";");
+ _mainPS.println(" ctx_spu[" + spuList.indexOf(spu)
+ + "].procContentsLen = " + spu.size() + ";");
+ }
+
+ _mainPS.println();
+
+ // Init the SPE control structure
+ _mainPS.println(" //Initiate SPEs control structure");
+ _mainPS.println(" int num = 0; ");
+ _mainPS.println(" for( num=0; num<NUM_SPES; num++){");
+ _mainPS.println(" data[num].argp = (void *)&(ctx_spu[num]);");
+ _mainPS.println(" }");
+ _mainPS.println();
+
+ _mainPS.println(" // Loop on all SPEs and for each perform three steps:");
+ _mainPS.println(" // - create SPE context");
+ _mainPS.println(" // - open images of SPE programs into main storage");
+ _mainPS.println(" // <spe_names> variable store the executable name");
+ _mainPS.println(" // - Load SPEs objects into SPE context local store");
+ _mainPS.println(" for( num=0; num<NUM_SPES; num++){");
+ _mainPS.println(" if ((data[num].spe_ctx = spe_context_create (SPE_MAP_PS, NULL)) == NULL) {");
+ _mainPS.println(" perror(\"Failed creating context\"); exit(1);");
+ _mainPS.println(" }");
+ _mainPS.println(" if (!(program[num] = spe_image_open(&spe_names[num][0]))) {");
+ _mainPS.println(" perror(\"Fail opening image\"); exit(1);");
+ _mainPS.println(" }");
+ _mainPS.println(" if (spe_program_load ( data[num].spe_ctx, program[num])) {");
+ _mainPS.println(" perror(\"Failed loading program\"); exit(1);");
+ _mainPS.println(" } ");
+ _mainPS.println(" }");
+ _mainPS.println("");
+
+ _mainPS.println(" // update the parameters of each SPE");
+ _mainPS.println(" for( num=0; num<NUM_SPES; num++){");
+ _mainPS.println(" if( (ea_ls_base[num] = (uint64_t)(spe_ls_area_get(data[num].spe_ctx))) == 0) {");
+ _mainPS.println(" perror(\"Failed map LS to main storage\"); exit(1);");
+ _mainPS.println(" }");
+ _mainPS.println(" ctx_spu[num].ea_base = ea_ls_base[num];");
+ _mainPS.println(" context_addr[num] = (uint64_t)data[num].spe_ctx;");
+ _mainPS.println(" }");
+
+ _mainPS.println(" for (num=0; num<NUM_SPES; num++) {");
+ _mainPS.println(" if ((mfc_ctl[num] = (spe_spu_control_area_t*) spe_ps_area_get(data[num].spe_ctx, SPE_CONTROL_AREA))==NULL){");
+ _mainPS.println(" perror(\"Failed mapping MFC control area\"); exit(1);");
+ _mainPS.println(" }");
+ _mainPS.println(" }");
+ _mainPS.println(" ");
+
+ _mainPS.println(" for (num=0; num<NUM_SPES; num++) {");
+ _mainPS.println(" ctx_spu[num].queueFromSPU = (uint64_t) queueFromSPU;");
+ _mainPS.println(" ctx_spu[num].queueOnSPU = (uint64_t) queueOnSPU;");
+ _mainPS.println(" ctx_spu[num].procContentsAll = (uint64_t) mfc_ctl;");
+ _mainPS.println(" ctx_spu[num].fifoTails = (uint64_t) fifoTails;");
+ _mainPS.println(" }");
+ _mainPS.println("");
+ _mainPS.println("// This is the PPU");
+ _mainPS.println(" ProcessData *ppu_Process_Wrapper = (ProcessData *)malloc(sizeof(ProcessData));");
+ _mainPS.println(" ppu_Process_Wrapper->procContentsAll = (uint64_t *) context_addr;");
+ _mainPS.println(" ppu_Process_Wrapper->queueFromSPU = (int32_t *)queueFromSPU;");
+ _mainPS.println(" ppu_Process_Wrapper->queueOnSPU = (int32_t *)queueOnSPU;");
+ _mainPS.println(" ppu_Process_Wrapper->ea_ls_base = (uint64_t *)ea_ls_base;");
+ _mainPS.println(" ppu_Process_Wrapper->fifoTails = (uint64_t *) fifoTails;");
+ _mainPS.println(" ");
+
+ _mainPS.println(" // create SPE pthreads");
+ _mainPS.println(" for( num=0; num<NUM_SPES; num++){");
+ _mainPS.println(" if(pthread_create(&data[num].pthread,NULL,&spu_pthread, &data[num])){");
+ _mainPS.println(" perror(\"Failed creating thread\"); exit(1);");
+ _mainPS.println(" }");
+ _mainPS.println(" }");
+ _mainPS.println(" ");
+ _mainPS.println(" // create PPE pthreads");
+ _mainPS.println(" pthread_t thread_ppu;");
+ _mainPS.println(" pthread_create( &thread_ppu, NULL, ppu, ppu_Process_Wrapper);");
+ _mainPS.println(" ");
+
+ _mainPS.println();
+ _mainPS.println(" // Loop on all SPEs and for each perform two steps:");
+ _mainPS.println(" // - wait for all the SPE pthread to complete");
+ _mainPS.println(" // - destroy the SPE contexts");
+ _mainPS.println(" for( num=0; num<NUM_SPES; num++){");
+ _mainPS.println(" if(pthread_join (data[num].pthread, NULL)) {");
+ _mainPS.println(" perror(\"Failed joining thread\"); exit (1);");
+ _mainPS.println(" }");
+ _mainPS.println(" ");
+ _mainPS.println(" if (spe_context_destroy( data[num].spe_ctx )) {");
+ _mainPS.println(" perror(\"Failed spe_context_destroy\"); exit(1);");
+ _mainPS.println(" }");
+ _mainPS.println(" }");
+ _mainPS.println(" ");
+ _mainPS.println(" // Join with PPU");
+ _mainPS.println(" if(pthread_join (thread_ppu, NULL)) {");
+ _mainPS.println(" perror(\"Failed joining thread\"); exit (1);");
+ _mainPS.println(" }");
+ _mainPS.println(" ");
+ _mainPS.println(" printf(\")PPE:) Complete running all super-fast SPEs and PPE\\n\");");
+ _mainPS.println("");
+
+ _mainPS.println(" return (0);");
+ _mainPS.println("}");
+ } catch (Exception e) {
+ System.out.println("CellModuleVisitor: exception occured: "
+ + e.getMessage());
+ e.printStackTrace();
+ }
+ }
+
+ /**
+ *
+ * @param x
+ * process that needs to be processed
+ */
+ public void visitComponent(Process x) {
+ }
+
+ /**
+ *
+ * @param x
+ * channel that needs to be processed
+ */
+ public void visitComponent(Channel x) {
+ }
+
+ /**
+ * Round the parameter to the next DMA-number up. Example: 23 gives 32.
+ *
+ * @param number
+ * number to round up
+ * @return next DMA-number
+ */
+ protected int roundDMA(int number) {
+ if (number > 16)
+ return number + 16 - (number % 16);
+ else if (number > 8)
+ return 16;
+ else if (number > 4)
+ return 8;
+ else if (number > 2)
+ return 4;
+ else if (number > 1)
+ return 2;
+ else
+ return 1;
+ }
+
+ protected CodePrintStream _mainPS = null;
+ protected String _dir = null;
+ protected HashMap<Port, Integer> _portMap;
+ protected CellMapping _mapping;
+
+}