Beispiel #1
0
static unsigned int __IndividualProblemLoad( Problem *thisprob,
                    unsigned int prob_i, Client *client, int *load_needed,
                    unsigned load_problem_count,
                    unsigned int *loaded_for_contest,
                    int *bufupd_pending )
{
  unsigned int did_load = 0;
  int retry_due_to_failed_loadstate = 0;

  TRACE_OUT((+1, "__IndivProbLoad()\n"));

  do /* while (retry_due_to_failed_loadstate) */
  {
    TRACE_OUT((0, "do /* while (retry_due_to_failed_loadstate) */\n"));
    WorkRecord wrdata;
    int update_on_current_contest_exhaust_flag = (client->connectoften & 4);
    long bufcount;
    int may_do_random_blocks, random_project = -1, proj_i;
    struct timeval tv;

    may_do_random_blocks = 0;
    for (proj_i = 0; proj_i < PROJECT_COUNT; ++proj_i)
    {
      int projectid = client->project_order_map[proj_i];
      if ((ProjectGetFlags(projectid) & PROJECTFLAG_RANDOM_BLOCKS) &&
          ((client->project_state[projectid] &
              (PROJECTSTATE_USER_DISABLED | PROJECTSTATE_CLOSED)) == 0))
      {
        may_do_random_blocks = 1;
        random_project = projectid;
        break;
      }
    }

    if (!may_do_random_blocks && client->last_buffupd_failed_time != 0
                              && CliClock(&tv) == 0)
    {
      // If no random work is allowed and if buffer updates are temporarily
      // disabled (retry delay not elapsed), then exit early to prevent the
      // client from doing unecessary file I/O operations until updates are
      // enabled.
      if (!tv.tv_sec)
        tv.tv_sec++;
      if (client->last_buffupd_failed_time + client->buffupd_retry_delay > tv.tv_sec) {
        *load_needed = NOLOAD_NORANDOM;
        return 0;   /* Retry delay not elapsed : did nothing */
      }
    }  

    retry_due_to_failed_loadstate = 0;
    bufcount = __loadapacket( client, &wrdata, 1, prob_i,
                              update_on_current_contest_exhaust_flag );

    if (bufcount < 0 && client->nonewblocks == 0)
    {
      if (CheckExitRequestTriggerNoIO())
        break;
      //Log("3. BufferUpdate(client,(BUFFERUPDATE_FETCH|BUFFERUPDATE_FLUSH),0)\n");
      TRACE_BUFFUPD((0, "BufferUpdate: reason = __IndividualProblemLoad && no more blocks\n"));
      int didupdate =
         BufferUpdate(client,(BUFFERUPDATE_FETCH|BUFFERUPDATE_FLUSH),0);
      if (CheckExitRequestTriggerNoIO())
        break;
      if (!(didupdate < 0))
      {
        if (didupdate!=0)
          *bufupd_pending&=~(didupdate&(BUFFERUPDATE_FLUSH|BUFFERUPDATE_FETCH));
        if ((didupdate & BUFFERUPDATE_FETCH) != 0) /* fetched successfully */
          bufcount = __loadapacket( client, &wrdata, 0, prob_i,
                                    update_on_current_contest_exhaust_flag );
      }
    }

    *load_needed = 0;
    if (bufcount >= 0) /* load from file suceeded */
      *load_needed = 0;
    else if (!may_do_random_blocks || client->blockcount < 0)
      *load_needed = NOLOAD_NORANDOM; /* -1 */
    else if (client->nonewblocks)
      *load_needed = NOLOAD_NONEWBLOCKS;
    else  /* using randoms is permitted */
      *load_needed = 0;

    TRACE_OUT((0, "bufcount = %ld, load_needed = %d, may_do_randoms = %d\n", bufcount, *load_needed, may_do_random_blocks));
    TRACE_BUFFUPD((0, "__Indiv...Load: bufcount = %ld, load_needed = %d\n", bufcount, *load_needed));
    if (*load_needed == 0)
    {
      u32 timeslice = 0x10000;
      int expected_cpu = 0, expected_core = 0;
      int expected_os  = 0, expected_build = 0;
      const ContestWork *work = &wrdata.work;
      int res = -1;

      #if (defined(INIT_TIMESLICE) && (INIT_TIMESLICE >= 64))
      timeslice = INIT_TIMESLICE;
      #endif

      if (bufcount < 0) /* normal load from buffer failed */
      {                 /* so generate random */
        work = CONTESTWORK_MAGIC_RANDOM;
        *loaded_for_contest = random_project;
      }
      else
      {
        *loaded_for_contest = (unsigned int)(wrdata.contest);
        // If a user gets a packet already started by another user,
        // force a restart to occur (may happen when the input buffer
        // is shared).
        if (strcmp(wrdata.id, "*****@*****.**") == 0
            || strncmp(wrdata.id, client->id, sizeof(wrdata.id)-1) == 0)
        {
          expected_cpu   = wrdata.cpu;
          expected_core  = wrdata.core;
          expected_os    = wrdata.os;
          expected_build = wrdata.build;
        }
        work = &wrdata.work;

        /* if the total number of packets in buffers is less than the number
           of crunchers running then post a fetch request. This means that the
           effective minimum threshold is always >= num crunchers
        */
        if (((unsigned long)(bufcount)) < (load_problem_count - prob_i))
        {
          *bufupd_pending |= BUFFERUPDATE_FETCH;
        }
      }

      /* loadstate can fail if it selcore fails or the previous problem */
      /* hadn't been purged, or the contest isn't available or ... */

      res = ProblemLoadState( thisprob, work, *loaded_for_contest, timeslice,
                    expected_cpu, expected_core, expected_os, expected_build, client );

      if (res != 0)
      {
        /* The problem with LoadState() failing is that it implicitely
        ** causes the block to be discarded, which means, that the
        ** keyserver network will reissue it - a senseless undertaking
        ** if the data itself is invalid.
        */
        if (res != -2)
        {
          retry_due_to_failed_loadstate = 1;
        }
        else
        {
          // should never happen
          Log("Serious ProblemLoadState() error! Aborting!\n");
          RaiseExitRequestTrigger();
        }
      }
      else
      {
        *load_needed = 0;
        did_load = 1;

        ClientEventSyncPost( CLIEVENT_PROBLEM_STARTED, &prob_i, sizeof(prob_i) );

        if (load_problem_count <= COMBINEMSG_THRESHOLD)
        {
          ProblemInfo info;
          if (ProblemGetInfo( thisprob, &info, P_INFO_S_PERMIL | P_INFO_SIGBUF   |
                                               P_INFO_DCOUNT   | P_INFO_EXACT_PE ) != -1)
          {
            const char *extramsg = "";
            char ddonebuf[15];
            char perdone[32];

            *loaded_for_contest = thisprob->pub_data.contest;
            if (thisprob->pub_data.was_reset)
              extramsg="\nPacket was from a different user/core/client cpu/os/build.";
            else if (info.s_permille > 0 && info.s_permille < 1000)
            {
              sprintf(perdone, " (%u.%u0%% done)", (info.s_permille/10), (info.s_permille%10));
              extramsg = perdone;
            }
            else if (info.dcounthi || info.dcountlo)
            {
              strcat( strcat( strcpy(perdone, " ("), U64stringify(ddonebuf, sizeof(ddonebuf),
                                                                  info.dcounthi, info.dcountlo,
                                                                  2, info.unit)),
                                                     " done)");
              extramsg = perdone;
            }

            if (load_problem_count > 1) {
              Log("%s #%c: Loaded %s%s%s\n", info.name, ProblemLetterId(prob_i),
                      ((thisprob->pub_data.is_random)?("random "):("")),
                      info.sigbuf, extramsg );
            }
            else {
              Log("%s: Loaded %s%s%s\n", info.name,
                      ((thisprob->pub_data.is_random)?("random "):("")),
                      info.sigbuf, extramsg );
            }
          } /* if (thisprob->GetProblemInfo(...) != -1) */
        } /* if (load_problem_count <= COMBINEMSG_THRESHOLD) */
      } /* if (LoadState(...) != -1) */
    } /* if (*load_needed == 0) */
  } while (retry_due_to_failed_loadstate);

  TRACE_OUT((-1, "__IndivProbLoad() => %d\n", did_load));
  return did_load;
}
Beispiel #2
0
static long __bench_or_test( Client *client, int which, 
                            unsigned int cont_i, unsigned int benchsecs, int in_corenum )
{
  long rc = -1;
  /* FIXME: without -devicenum, test/bench will be run on GPU 0 only */
  int device = hackGetUsedDeviceIndex(client, 0);

  if (selcore_initlev > 0                  /* core table is initialized? */
      && cont_i < CONTEST_COUNT)           /* valid contest id? */
  {
    /* save current state */
    int user_cputype = selcorestatics.user_cputype[cont_i];
    int corenum = selcorestatics.corenum[cont_i];
    int coreidx, corecount = corecount_for_contest( cont_i );
    int fastest = -1;
    int hardcoded = selcoreGetPreselectedCoreForProject(cont_i, device);
    u32 bestrate_hi = 0, bestrate_lo = 0, refrate_hi = 0, refrate_lo = 0;

    rc = 0; /* assume nothing done */
    for (coreidx = 0; coreidx < corecount; coreidx++)
    {
      /* only bench/test cores that won't be automatically substituted */
      if (apply_selcore_substitution_rules(cont_i, coreidx, device) == coreidx)
      {
        if (in_corenum < 0)
            selcorestatics.user_cputype[cont_i] = coreidx; /* as if user set it */
        else
        {
          if( in_corenum < corecount )
          {
            selcorestatics.user_cputype[cont_i] = in_corenum;
            coreidx = corecount;
          }
          else  /* invalid core selection, test them all */
          {
            selcorestatics.user_cputype[cont_i] = coreidx;
            in_corenum = -1;
          }
        }
        selcorestatics.corenum[cont_i] = -1; /* reset to show name */

        if (which == 't') /* selftest */
          rc = SelfTest( client, cont_i );
        else if (which == 's') /* stresstest */
          rc = StressTest( client, cont_i );
        else {
          u32 temprate_hi, temprate_lo;
          rc = TBenchmark( client, cont_i, benchsecs, 0, &temprate_hi, &temprate_lo );
          if (rc > 0 && selcorestatics.corenum[cont_i] == hardcoded) {
            refrate_hi = temprate_hi;
            refrate_lo = temprate_lo;
          }
          if (rc > 0 && (temprate_hi > bestrate_hi || (temprate_hi == bestrate_hi && temprate_lo > bestrate_lo))) {
            bestrate_hi = temprate_hi;
            bestrate_lo = temprate_lo;
            fastest     = selcorestatics.corenum[cont_i];
          }
        }
        #if (CLIENT_OS != OS_WIN32 || !defined(SMC))
        if (rc <= 0) /* failed (<0) or not supported (0) */
          break; /* stop */
        #else
        // HACK! to ignore failed benchmark for x86 rc5 smc core #7 if
        // started from menu and another cruncher is active in background.
        if (rc <= 0) /* failed (<0) or not supported (0) */
        {
          if ( which == 'b' &&  cont_i == RC5 && coreidx == 7 )
            ; /* continue */
          else
            break; /* stop */
        }
        #endif
      }
    } /* for (coreidx = 0; coreidx < corecount; coreidx++) */

    selcorestatics.user_cputype[cont_i] = user_cputype;
    selcorestatics.corenum[cont_i] = corenum;

    /* Summarize the results if multiple cores have been benchmarked (#4108) */
#if (CLIENT_CPU != CPU_CELLBE)
    /* Not applicable for Cell due to PPU/SPU core selection hacks */
    if (in_corenum < 0 && fastest >= 0 && (bestrate_hi != 0 || bestrate_lo != 0)) {
      double percent = 100.0 * ((double)refrate_hi  * 4294967296.0 + (double)refrate_lo) /
                               ((double)bestrate_hi * 4294967296.0 + (double)bestrate_lo);
      char bestrate_str[32], refrate_str[32];

      U64stringify(bestrate_str, sizeof(bestrate_str), bestrate_hi, bestrate_lo, 2, CliGetContestUnitFromID(cont_i));
      U64stringify(refrate_str,  sizeof(refrate_str),  refrate_hi,  refrate_lo,  2, CliGetContestUnitFromID(cont_i));

      Log("%s benchmark summary :\n"
          "Default core : #%d (%s) %s/sec\n"
          "Fastest core : #%d (%s) %s/sec\n",
          CliGetContestNameFromID(cont_i), hardcoded,
          (hardcoded >= 0 ? selcoreGetDisplayName(cont_i, hardcoded) : "undefined"),
          refrate_str,
          fastest, selcoreGetDisplayName(cont_i, fastest), bestrate_str);
          
      if (percent < 100 && hardcoded >= 0 && hardcoded != fastest) {
        if (percent >= 97) {
          Log("Core #%d is marginally faster than the default core.\n"
              "Testing variability might lead to pick one or the other.\n",
              fastest);
        }
        else {
          Log("Core #%d is significantly faster than the default core.\n"
#if (CLIENT_CPU != CPU_CUDA && CLIENT_CPU != CPU_ATI_STREAM && CLIENT_CPU != CPU_OPENCL)
              "Please file a bug report along with the output of\n-cpuinfo.\n"
#else
              "The GPU core selection has been made as a tradeoff between core speed\n"
              "and responsiveness of the graphical desktop.\n"
              "Please file a bug report along with the output of -gpuinfo\n"
              "only if the the faster core selection does not degrade graphics performance.\n"
#endif
              "Changes in cores and selection are frequently made,\n"
              "so be sure to test with the latest client version,\n"
              "typically a pre-release, before filing a bug report.\n",
              fastest);
        }
      }
    }
#endif // CPU_CELLBE

#if (CLIENT_OS == OS_RISCOS) && defined(HAVE_X86_CARD_SUPPORT)
    if (rc > 0 && cont_i == RC5 && 
          GetNumberOfDetectedProcessors() > 1) /* have x86 card */
    {
      Problem *prob = ProblemAlloc(); /* so bench/test gets threadnum+1 */ /* FIXME: not true anymore */
      rc = -1; /* assume alloc failed */
      if (prob)
      {
        Log("RC5: using x86 core.\n" );
        if (which != 's') /* bench */
          rc = TBenchmark( client, cont_i, benchsecs, 0 );
        else
          rc = SelfTest( client, cont_i );
        ProblemFree(prob);
      }
    }
    #endif

  } /* if (cont_i < CONTEST_COUNT) */

  return rc;
}
Beispiel #3
0
static unsigned int __IndividualProblemSave( Problem *thisprob,
                unsigned int prob_i, Client *client, int *is_empty,
                unsigned load_problem_count, unsigned int *contest,
                int *bufupd_pending, int unconditional_unload,
                int abortive_action )
{
  unsigned int did_save = 0;

  DNETC_UNUSED_PARAM(prob_i);

  *contest = 0;
  *is_empty = 1; /* assume not initialized */
  if ( ProblemIsInitialized(thisprob) )
  {
    WorkRecord wrdata;
    int resultcode;
    unsigned int cont_i;
    memset( (void *)&wrdata, 0, sizeof(WorkRecord));
    resultcode = ProblemRetrieveState( thisprob, &wrdata.work, &cont_i, 0, 0 );

    *is_empty = 0; /* assume problem is in use */

    if (resultcode == RESULT_FOUND || resultcode == RESULT_NOTHING ||
       unconditional_unload || resultcode < 0 /* core error */ ||
      (thisprob->pub_data.loaderflags & (PROBLDR_DISCARD|PROBLDR_FORCEUNLOAD)) != 0)
    {
      int finito = (resultcode==RESULT_FOUND || resultcode==RESULT_NOTHING);
      const char *action_msg = 0;
      const char *reason_msg = 0;
      int discarded = 0;
      char ratebuf[32];
      char dcountbuf[64]; /* we use this as scratch space too */
      struct timeval tv;
      ProblemInfo info;
      info.rate.ratebuf = ratebuf;
      info.rate.size = sizeof(ratebuf);

      *contest = cont_i;
      *is_empty = 1; /* will soon be */

      wrdata.contest    = cont_i;
      wrdata.resultcode = resultcode;
      wrdata.cpu        = FILEENTRY_CPU(thisprob->pub_data.client_cpu);
      wrdata.os         = FILEENTRY_OS;
      wrdata.build      = FILEENTRY_BUILD;
      wrdata.core       = FILEENTRY_CORE(thisprob->pub_data.coresel);
      strncpy( wrdata.id, client->id , sizeof(wrdata.id));
      wrdata.id[sizeof(wrdata.id)-1]=0;

      if (finito)
      {
        wrdata.os      = CLIENT_OS;
        #if (CLIENT_OS == OS_RISCOS)
        if (prob_i == 1)
          wrdata.cpu   = CPU_X86;
        else
        #endif
        wrdata.cpu     = CLIENT_CPU;
        wrdata.build   = CLIENT_VERSION;
        wrdata.core    = FILEENTRY_CORE(thisprob->pub_data.coresel);
        ClientEventSyncPost( CLIEVENT_PROBLEM_FINISHED, &prob_i, sizeof(prob_i) );
      }

      if ((thisprob->pub_data.loaderflags & PROBLDR_DISCARD)!=0)
      {
        action_msg = "Discarded";
        reason_msg = "project disabled/closed";
        discarded = 1;
      }
      else if (resultcode < 0)
      {
        action_msg = "Discarded";
        reason_msg = "core error";
        discarded = 1;
      }
      else if (PutBufferRecord( client, &wrdata ) < 0)
      {
        action_msg = "Discarded";
        reason_msg = "buffer error - unable to save";
        discarded = 1;
      }
      else
      {
        did_save = 1;
        if (client->nodiskbuffers)
          *bufupd_pending |= BUFFERUPDATE_FLUSH;
        if (__check_outbufthresh_limit( client, cont_i, -1, 0,bufupd_pending))
        { /* adjust bufupd_pending if outthresh has been crossed */
          //Log("1. *bufupd_pending |= BUFFERUPDATE_FLUSH;\n");
        }

        if (load_problem_count > COMBINEMSG_THRESHOLD)
          ; /* nothing */
        else if (thisprob->pub_data.was_truncated)
        {
          action_msg = "Skipped";
          discarded  = 1;
          reason_msg = thisprob->pub_data.was_truncated;
        }
        else if (!finito)
          action_msg = "Saved";
        else
          action_msg = "Completed";
      }
      if (ProblemGetInfo( thisprob, &info, P_INFO_E_TIME   | P_INFO_SWUCOUNT | P_INFO_C_PERMIL |
                                           P_INFO_SIGBUF   | P_INFO_RATEBUF  | P_INFO_TCOUNT   |
                                           P_INFO_CCOUNT   | P_INFO_DCOUNT   | P_INFO_EXACT_PE ) != -1)
      {
        char info_name[32];
        tv.tv_sec = info.elapsed_secs; tv.tv_usec = info.elapsed_usecs;

        if (load_problem_count > 1) {
          sprintf(info_name, "%s #%c", info.name, ProblemLetterId(prob_i));
        }
        else
          strcpy(info_name, info.name);

        if (finito && !discarded && !info.is_test_packet)
          CliAddContestInfoSummaryData(cont_i,info.ccounthi,info.ccountlo,&tv,info.swucount);

        if (action_msg)
        {
          if (reason_msg) /* was discarded */
          {
            //[....] Discarded CSC 12345678:ABCDEF00 4*2^28
            //       (project disabled/closed)
            Log("%s: %s %s%c(%s)\n", info_name, action_msg, info.sigbuf,
                       ((strlen(reason_msg)>10)?('\n'):(' ')), reason_msg );
          }
          else
          {
            U64stringify(dcountbuf, (15<sizeof(dcountbuf))?15:sizeof(dcountbuf),
                         info.dcounthi, info.dcountlo, 2, info.unit);
            if (finito && info.is_test_packet) /* finished test packet */
              strcat( strcpy( dcountbuf,"Test: RESULT_"),
                     ((resultcode==RESULT_NOTHING)?("NOTHING"):("FOUND")) );
            else if (finito) /* finished non-test packet */
            {
              char *p = strrchr(info.sigbuf,':'); /* HACK! to supress too long */
              if (p) *p = '\0';            /* crypto "Completed" lines */
              sprintf( dcountbuf, "%u.%02u stats units",
                                  info.swucount/100, info.swucount%100);
            }
            else if (info.c_permille > 0)
              sprintf( dcountbuf, "%u.%u0%% done", info.c_permille/10, info.c_permille%10);
            else
              strcat( dcountbuf, " done" );

            //[....] RC5: Saved 12345678:ABCDEF00 4*2^28 (5.20% done)
            //       1.23:45:67:89 - [987,654,321 keys/s]
            //[....] OGR: Saved 25/1-6-13-8-16-18 (12.34 Mnodes done)
            //       1.23:45:67:89 - [987,654,321 nodes/s]
            //[....] RC5: Completed 68E0D85A:A0000000 4*2^28 (4.00 stats units)
            //       1.23:45:67:89 - [987,654,321 keys/s]
            //[....] OGR: Completed 22/1-3-5-7 (12.30 stats units)
            //       1.23:45:67:89 - [987,654,321 nodes/s]
            //[....] OGR: 25/1-2-4-5-8-10 [12,345,578,910 nodes]
            Log("%s: %s %s (%s)\n%s - [%s/s]\n",
              info_name, action_msg, info.sigbuf, dcountbuf,
              CliGetTimeString( &tv, 2 ), info.rate.ratebuf );
            if (finito && info.show_exact_iterations_done)
            {
              Log("%s: %s [%s]\n", info_name, info.sigbuf,
                  ProblemComputeRate(cont_i, 0, 0, info.tcounthi, info.tcountlo,
                                     0, 0, dcountbuf, sizeof(dcountbuf)));
            }
          } /* if (reason_msg) else */
        } /* if (action_msg) */
      } /* if (thisprob->GetProblemInfo( ... ) != -1) */

      /* we can purge the object now */
      /* we don't wait when aborting. thread might be hung */
      ProblemRetrieveState( thisprob, NULL, NULL, 1, abortive_action /*==dontwait*/ );

    } /* unload needed */
  } /* is initialized */

  return did_save;
}