s5z / zsim

A fast and scalable x86-64 multicore simulator
GNU General Public License v2.0
328 stars 183 forks source link

Running OpenMP benchmark is causing stall's #244

Open avacoder42 opened 4 years ago

avacoder42 commented 4 years ago

Hi, I'm running zsim application in ramulator with PIM-hooks and ROI-hooks for Polybench OpenMP applications like Triangular matrix multiplication(trmm) , Cholesky decomposition. For all dimensions from small to large datasets, I get the following warning message and the run-time is more than an hour, I'm questioning if these warning's need to be addressed and corrected? Thank you

[S 0] WARN: Futex wake matching failed (0/31) (external/ff waiters?
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 30 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 30 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 30 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 30 secs so far
[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 30 secs so far
[H] WARN: Stalled for 40 secs so far

[H] WARN: Stalled for 20 secs so far
[H] WARN: Stalled for 30 secs so far
[H] WARN: Stalled for 20 secs so far
[S 0] Detected possible stall due to fake leaves (1 current)
[S 0]  [0/0] futex (202) @ 0x7fffe2e08111
[S 0] Blacklisting from future fake leaves: [0] futex @ 0x7fffe2e08111 | arg0 0x555555757444 arg1 0x80

Configuration is as follows -

// This system is similar to a 6-core, 2.4GHz Westmere with 10 Niagara-like cores attached to the L3
sys = {
    lineSize = 64;
    frequency = 2400;

    cores = {
        core = {
            type = "OOO";
            cores = 32;
            icache = "l1i";
            dcache = "l1d";
        };
    };

    caches = {
        l1d = {
            array = {
                type = "SetAssoc";
                ways = 8;
            };
            caches = 32;
            latency = 4;
            size = 32768;
        };
        l1i = {
            array = {
                type = "SetAssoc";
                ways = 4;
            };
            caches = 32;
            latency = 3;
            size = 32768;
        };
        l2 = {
            array = {
                type = "SetAssoc";
                ways = 8;
            };
        //type = "Timing";
        //mshrs = 10;
            caches = 32;
            latency = 7;
            children = "l1i|l1d";
            size = 262144;
        };
        l3 = {
            array = {
                hash = "H3";
                type = "SetAssoc";
                ways = 16;
            };
        //type = "Timing";
        //mshrs = 16;
            banks = 32;
            caches = 1;
            latency = 27;
            children = "l2";
        size = 67108864;
        };

    };

    mem = {
        type = "Traces";
        instr_traces = true;
          only_offload = true;
          pim_traces = true;

        outFile = "pim-poly_cholesky_32.out"
    };

};

sim = {
    phaseLength = 10000;
    maxTotalInstrs = 10000000000L;
    statsPhaseInterval = 1000;
    printHierarchy = true;
    // attachDebugger = True;
};

process0 = {
    command = "benchmarks/PolyBench-ACC-master/OpenMP/linear-algebra/kernels/cholesky/cholesky" ;
    startFastForwarded = True;
//    command = "ls -la";
//    command = "unzip tracesLois.out.gz";
};

And code with hooks -

static
void kernel_cholesky(int n,
             DATA_TYPE POLYBENCH_1D(p,N,n),
             DATA_TYPE POLYBENCH_2D(A,N,N,n,n))
{

  int i, j, k;
  int    num_omp_threads;
    num_omp_threads = 32;
  DATA_TYPE x;
  #pragma scop
  #pragma omp parallel
  {  

    #pragma omp for private (j,k)
    for (i = 0; i < _PB_N; ++i)
      { 
        zsim_PIM_function_begin();
          x = A[i][i];
          for (j = 0; j <= i - 1; ++j)

              x = x - A[i][j] * A[i][j];
            p[i] = 1.0 /sqrt(x);

          for (j = i + 1; j < _PB_N; ++j)
            {
              x = A[i][j];
              for (k = 0; k <= i - 1; ++k)
                x = x - A[j][k] * A[i][k];
                A[j][i] = x * p[i];
            }
          zsim_PIM_function_end(); 
      }

  }
  #pragma endscop

}

int main(int argc, char** argv)
{
  /* Retrieve problem size. */
  int n = N;

  /* Variable declaration/allocation. */
  POLYBENCH_2D_ARRAY_DECL(A, DATA_TYPE, N, N, n, n);
  POLYBENCH_1D_ARRAY_DECL(p, DATA_TYPE, N, n);

  /* Initialize array(s). */
  init_array (n, POLYBENCH_ARRAY(p), POLYBENCH_ARRAY(A));

  /* Start timer. */
  polybench_start_instruments;

  /* Run kernel. */
  zsim_roi_begin();
  kernel_cholesky (n, POLYBENCH_ARRAY(p), POLYBENCH_ARRAY(A));
  zsim_roi_end();
  /* Stop and print timer. */
  polybench_stop_instruments;
  polybench_print_instruments;

  /* Prevent dead-code elimination. All live-out data must be printed
     by the function call in argument. */
  polybench_prevent_dce(print_array(n, POLYBENCH_ARRAY(A)));

  /* Be clean. */
  POLYBENCH_FREE_ARRAY(A);
  POLYBENCH_FREE_ARRAY(p);

  return 0;
}