gcc/libgomp/testsuite/libgomp.oacc-c-c++-common/mode-transitions.c

/* Miscellaneous test cases for gang/worker/vector mode transitions.  */

#include <assert.h>
#include <stdbool.h>
#include <stdlib.h>
#include <math.h>
#include <openacc.h>


/* Test basic vector-partitioned mode transitions.  */

void t1()
{
  int n = 0, arr[32], i;

  for (i = 0; i < 32; i++)
    arr[i] = 0;

  #pragma acc parallel copy(n, arr) \
		       num_gangs(1) num_workers(1) vector_length(32)
  {
    int j;
    n++;
    #pragma acc loop vector
    for (j = 0; j < 32; j++)
      arr[j]++;
    n++;
  }

  assert (n == 2);

  for (i = 0; i < 32; i++)
    assert (arr[i] == 1);
}


/* Test vector-partitioned, gang-partitioned mode.  */

void t2()
{
  int n[32], arr[1024], i;
  
  for (i = 0; i < 1024; i++)
    arr[i] = 0;

  for (i = 0; i < 32; i++)
    n[i] = 0;

  #pragma acc parallel copy(n, arr) \
		       num_gangs(32) num_workers(1) vector_length(32)
  {
    int j, k;

    #pragma acc loop gang(static:*)
    for (j = 0; j < 32; j++)
      n[j]++;

    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      #pragma acc loop vector
      for (k = 0; k < 32; k++)
	arr[j * 32 + k]++;

    #pragma acc loop gang(static:*)
    for (j = 0; j < 32; j++)
      n[j]++;
  }

  for (i = 0; i < 32; i++)
    assert (n[i] == 2);

  for (i = 0; i < 1024; i++)
    assert (arr[i] == 1);
}


/* Test conditions inside vector-partitioned loops.  */

void t4()
{
  int n[32], arr[1024], i;

  for (i = 0; i < 1024; i++)
    arr[i] = i;

  for (i = 0; i < 32; i++)
    n[i] = 0;

  #pragma acc parallel copy(n, arr) \
		       num_gangs(32) num_workers(1) vector_length(32)
  {
    int j, k;

    #pragma acc loop gang(static:*)
    for (j = 0; j < 32; j++)
      n[j]++;

    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      {
	#pragma acc loop vector
	for (k = 0; k < 32; k++)
	  if ((arr[j * 32 + k] % 2) != 0)
	    arr[j * 32 + k] *= 2;
      }

    #pragma acc loop gang(static:*)
    for (j = 0; j < 32; j++)
      n[j]++;
  }

  for (i = 0; i < 32; i++)
    assert (n[i] == 2);

  for (i = 0; i < 1024; i++)
    assert (arr[i] == ((i % 2) == 0 ? i : i * 2));
}


/* Test conditions inside gang-partitioned/vector-partitioned loops.  */

void t5()
{
  int n[32], arr[1024], i;

  for (i = 0; i < 1024; i++)
    arr[i] = i;

  for (i = 0; i < 32; i++)
    n[i] = 0;

  #pragma acc parallel copy(n, arr) \
		       num_gangs(32) num_workers(1) vector_length(32)
  {
    int j;

    #pragma acc loop gang(static:*)
    for (j = 0; j < 32; j++)
      n[j]++;

    #pragma acc loop gang vector
    for (j = 0; j < 1024; j++)
      if ((arr[j] % 2) != 0)
	arr[j] *= 2;

    #pragma acc loop gang(static:*)
    for (j = 0; j < 32; j++)
      n[j]++;
  }

  for (i = 0; i < 32; i++)
    assert (n[i] == 2);

  for (i = 0; i < 1024; i++)
    assert (arr[i] == ((i % 2) == 0 ? i : i * 2));
}


/* Test trivial operation of vector-single mode.  */

void t7()
{
  int n = 0;
  #pragma acc parallel copy(n) \
		       num_gangs(1) num_workers(1) vector_length(32)
  {
    n++;
  }
  assert (n == 1);
}


/* Test vector-single, gang-partitioned mode.  */

void t8()
{
  int arr[1024];
  int gangs;

  for (gangs = 1; gangs <= 1024; gangs <<= 1)
    {
      int i;

      for (i = 0; i < 1024; i++)
	arr[i] = 0;

      #pragma acc parallel copy(arr) \
			   num_gangs(gangs) num_workers(1) vector_length(32)
      {
	int j;
	#pragma acc loop gang
	for (j = 0; j < 1024; j++)
	  arr[j]++;
      }

      for (i = 0; i < 1024; i++)
	assert (arr[i] == 1);
    }
}


/* Test conditions in vector-single mode.  */

void t9()
{
  int arr[1024];
  int gangs;

  for (gangs = 1; gangs <= 1024; gangs <<= 1)
    {
      int i;

      for (i = 0; i < 1024; i++)
	arr[i] = 0;

      #pragma acc parallel copy(arr) \
			   num_gangs(gangs) num_workers(1) vector_length(32)
      {
	int j;
	#pragma acc loop gang
	for (j = 0; j < 1024; j++)
	  if ((j % 3) == 0)
	    arr[j]++;
	  else
	    arr[j] += 2;
      }

      for (i = 0; i < 1024; i++)
	assert (arr[i] == ((i % 3) == 0) ? 1 : 2);
    }
}


/* Test switch in vector-single mode.  */

void t10()
{
  int arr[1024];
  int gangs;

  for (gangs = 1; gangs <= 1024; gangs <<= 1)
    {
      int i;

      for (i = 0; i < 1024; i++)
	arr[i] = 0;

      #pragma acc parallel copy(arr) \
			   num_gangs(gangs) num_workers(1) vector_length(32)
      {
	int j;
	#pragma acc loop gang
	for (j = 0; j < 1024; j++)
	  switch (j % 5)
	    {
	    case 0: arr[j] += 1; break;
	    case 1: arr[j] += 2; break;
	    case 2: arr[j] += 3; break;
	    case 3: arr[j] += 4; break;
	    case 4: arr[j] += 5; break;
	    default: arr[j] += 99;
	    }
      }

      for (i = 0; i < 1024; i++)
	assert (arr[i] == (i % 5) + 1);
    }
}


/* Test switch in vector-single mode, initialise array on device.  */

void t11()
{
  int arr[1024];
  int i;

  for (i = 0; i < 1024; i++)
    arr[i] = 99;

  #pragma acc parallel copy(arr) \
		       num_gangs(1024) num_workers(1) vector_length(32)
  {
    int j;

    /* This loop and the one following must be distributed to available gangs
       in the same way to ensure data dependencies are not violated (hence the
       "static" clauses).  */
    #pragma acc loop gang(static:*)
    for (j = 0; j < 1024; j++)
      arr[j] = 0;
    
    #pragma acc loop gang(static:*)
    for (j = 0; j < 1024; j++)
      switch (j % 5)
	{
	case 0: arr[j] += 1; break;
	case 1: arr[j] += 2; break;
	case 2: arr[j] += 3; break;
	case 3: arr[j] += 4; break;
	case 4: arr[j] += 5; break;
	default: arr[j] += 99;
	}
  }

  for (i = 0; i < 1024; i++)
    assert (arr[i] == (i % 5) + 1);
}


/* Test multiple conditions in vector-single mode.  */

#define NUM_GANGS 4096
void t12()
{
  bool fizz[NUM_GANGS], buzz[NUM_GANGS], fizzbuzz[NUM_GANGS];
  int i;

  #pragma acc parallel copyout(fizz, buzz, fizzbuzz) \
		       num_gangs(NUM_GANGS) num_workers(1) vector_length(32)
  {
    int j;
    
    /* This loop and the one following must be distributed to available gangs
       in the same way to ensure data dependencies are not violated (hence the
       "static" clauses).  */
    #pragma acc loop gang(static:*)
    for (j = 0; j < NUM_GANGS; j++)
      fizz[j] = buzz[j] = fizzbuzz[j] = 0;
    
    #pragma acc loop gang(static:*)
    for (j = 0; j < NUM_GANGS; j++)
      {
	if ((j % 3) == 0 && (j % 5) == 0)
	  fizzbuzz[j] = 1;
	else
	  {
	    if ((j % 3) == 0)
	      fizz[j] = 1;
	    else if ((j % 5) == 0)
	      buzz[j] = 1;
	  }
      }
  }

  for (i = 0; i < NUM_GANGS; i++)
    {
      assert (fizzbuzz[i] == ((i % 3) == 0 && (i % 5) == 0));
      assert (fizz[i] == ((i % 3) == 0 && (i % 5) != 0));
      assert (buzz[i] == ((i % 3) != 0 && (i % 5) == 0));
    }
}
#undef NUM_GANGS


/* Test worker-partitioned/vector-single mode.  */

void t13()
{
  int arr[32 * 8], i;

  for (i = 0; i < 32 * 8; i++)
    arr[i] = 0;

  #pragma acc parallel copy(arr) \
		       num_gangs(8) num_workers(8) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      {
	int k;
	#pragma acc loop worker
	for (k = 0; k < 8; k++)
          arr[j * 8 + k] += j * 8 + k;
      }
  }

  for (i = 0; i < 32 * 8; i++)
    assert (arr[i] == i);
}


/* Test worker-single/worker-partitioned transitions.  */

void t16()
{
  int n[32], arr[32 * 32], i;

  for (i = 0; i < 32 * 32; i++)
    arr[i] = 0;

  for (i = 0; i < 32; i++)
    n[i] = 0;

  #pragma acc parallel copy(n, arr) \
		       num_gangs(8) num_workers(16) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      {
	int k;

	n[j]++;

	#pragma acc loop worker
	for (k = 0; k < 32; k++)
          arr[j * 32 + k]++;

	n[j]++;

	#pragma acc loop worker
	for (k = 0; k < 32; k++)
          arr[j * 32 + k]++;

	n[j]++;

	#pragma acc loop worker
	for (k = 0; k < 32; k++)
          arr[j * 32 + k]++;

	n[j]++;
      }
  }

  for (i = 0; i < 32; i++)
    assert (n[i] == 4);

  for (i = 0; i < 32 * 32; i++)
    assert (arr[i] == 3);
}


/* Test correct synchronisation between worker-partitioned loops.  */

void t17()
{
  int arr_a[32 * 32], arr_b[32 * 32], i;
  int num_workers, num_gangs;

  for (num_workers = 1; num_workers <= 32; num_workers <<= 1)
    for (num_gangs = 1; num_gangs <= 32; num_gangs <<= 1)
      {
	for (i = 0; i < 32 * 32; i++)
	  arr_a[i] = i;

	#pragma acc parallel copyin(arr_a) copyout(arr_b) \
			     num_gangs(num_gangs) num_workers(num_workers) vector_length(32)
	{
	  int j;
	  #pragma acc loop gang
	  for (j = 0; j < 32; j++)
	    {
	      int k;

	      #pragma acc loop worker
	      for (k = 0; k < 32; k++)
        	arr_b[j * 32 + (31 - k)] = arr_a[j * 32 + k] * 2;

	      #pragma acc loop worker
	      for (k = 0; k < 32; k++)
        	arr_a[j * 32 + (31 - k)] = arr_b[j * 32 + k] * 2;

	      #pragma acc loop worker
	      for (k = 0; k < 32; k++)
        	arr_b[j * 32 + (31 - k)] = arr_a[j * 32 + k] * 2;
	    }
	}

	for (i = 0; i < 32 * 32; i++)
	  assert (arr_b[i] == (i ^ 31) * 8);
      }
}


/* Test correct synchronisation between worker+vector-partitioned loops.  */

void t18()
{
  int arr_a[32 * 32 * 32], arr_b[32 * 32 * 32], i;
  int num_workers, num_gangs;

  for (num_workers = 1; num_workers <= 32; num_workers <<= 1)
    for (num_gangs = 1; num_gangs <= 32; num_gangs <<= 1)
      {
	for (i = 0; i < 32 * 32 * 32; i++)
	  arr_a[i] = i;

	#pragma acc parallel copyin(arr_a) copyout(arr_b) \
			     num_gangs(num_gangs) num_workers(num_workers) vector_length(32)
	{
	  int j;
	  #pragma acc loop gang
	  for (j = 0; j < 32; j++)
	    {
	      int k;

	      #pragma acc loop worker vector
	      for (k = 0; k < 32 * 32; k++)
        	arr_b[j * 32 * 32 + (1023 - k)] = arr_a[j * 32 * 32 + k] * 2;

	      #pragma acc loop worker vector
	      for (k = 0; k < 32 * 32; k++)
        	arr_a[j * 32 * 32 + (1023 - k)] = arr_b[j * 32 * 32 + k] * 2;

	      #pragma acc loop worker vector
	      for (k = 0; k < 32 * 32; k++)
        	arr_b[j * 32 * 32 + (1023 - k)] = arr_a[j * 32 * 32 + k] * 2;
	    }
	}

	for (i = 0; i < 32 * 32 * 32; i++)
	  assert (arr_b[i] == (i ^ 1023) * 8);
      }
}


/* Test correct synchronisation between vector-partitioned loops in
   worker-partitioned mode.  */

void t19()
{
  int n[32 * 32], arr_a[32 * 32 * 32], arr_b[32 * 32 * 32], i;
  int num_workers, num_gangs;

  for (num_workers = 1; num_workers <= 32; num_workers <<= 1)
    for (num_gangs = 1; num_gangs <= 32; num_gangs <<= 1)
      {
	for (i = 0; i < 32 * 32 * 32; i++)
	  arr_a[i] = i;

	for (i = 0; i < 32 * 32; i++)
          n[i] = 0;

	#pragma acc parallel copy (n) copyin(arr_a) copyout(arr_b) \
			     num_gangs(num_gangs) num_workers(num_workers) vector_length(32)
	{
	  int j;
	  #pragma acc loop gang
	  for (j = 0; j < 32; j++)
	    {
	      int k;

	      #pragma acc loop worker
	      for (k = 0; k < 32; k++)
		{
		  int m;

		  n[j * 32 + k]++;

		  #pragma acc loop vector
		  for (m = 0; m < 32; m++)
		    {
	              if (((j * 1024 + k * 32 + m) % 2) == 0)
			arr_b[j * 1024 + k * 32 + (31 - m)]
			  = arr_a[j * 1024 + k * 32 + m] * 2;
		      else
			arr_b[j * 1024 + k * 32 + (31 - m)]
			  = arr_a[j * 1024 + k * 32 + m] * 3;
		    }

		  /* Test returning to vector-single mode...  */
		  n[j * 32 + k]++;

		  #pragma acc loop vector
		  for (m = 0; m < 32; m++)
		    {
	              if (((j * 1024 + k * 32 + m) % 3) == 0)
			arr_a[j * 1024 + k * 32 + (31 - m)]
			  = arr_b[j * 1024 + k * 32 + m] * 5;
		      else
			arr_a[j * 1024 + k * 32 + (31 - m)]
			  = arr_b[j * 1024 + k * 32 + m] * 7;
		    }

		  /* ...and back-to-back vector loops.  */

		  #pragma acc loop vector
		  for (m = 0; m < 32; m++)
		    {
	              if (((j * 1024 + k * 32 + m) % 2) == 0)
			arr_b[j * 1024 + k * 32 + (31 - m)]
			  = arr_a[j * 1024 + k * 32 + m] * 3;
		      else
			arr_b[j * 1024 + k * 32 + (31 - m)]
			  = arr_a[j * 1024 + k * 32 + m] * 2;
		    }
		}
	    }
	}

	for (i = 0; i < 32 * 32; i++)
          assert (n[i] == 2);

	for (i = 0; i < 32 * 32 * 32; i++)
          {
	    int m = 6 * ((i % 3) == 0 ? 5 : 7);
	    assert (arr_b[i] == (i ^ 31) * m);
	  }
      }
}


/* With -O0, variables are on the stack, not in registers.  Check that worker
   state propagation handles the stack frame.  */

void t20()
{
  int w0 = 0;
  int w1 = 0;
  int w2 = 0;
  int w3 = 0;
  int w4 = 0;
  int w5 = 0;
  int w6 = 0;
  int w7 = 0;

  int i;

#pragma acc parallel copy (w0, w1, w2, w3, w4, w5, w6, w7) \
		     num_gangs (1) num_workers (8)
  {
    int internal = 100;

#pragma acc loop worker
    for (i = 0; i < 8; i++)
      {
	switch (i)
	  {
	  case 0: w0 = internal; break;
	  case 1: w1 = internal; break;
	  case 2: w2 = internal; break;
	  case 3: w3 = internal; break;
	  case 4: w4 = internal; break;
	  case 5: w5 = internal; break;
	  case 6: w6 = internal; break;
	  case 7: w7 = internal; break;
	  default: break;
	  }
      }
  }

  if (w0 != 100
      || w1 != 100
      || w2 != 100
      || w3 != 100
      || w4 != 100
      || w5 != 100
      || w6 != 100
      || w7 != 100)
    __builtin_abort ();
}


/* Test worker-single/vector-single mode.  */

void t21()
{
  int arr[32], i;

  for (i = 0; i < 32; i++)
    arr[i] = 0;

  #pragma acc parallel copy(arr) \
		       num_gangs(8) num_workers(8) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      arr[j]++;
  }

  for (i = 0; i < 32; i++)
    assert (arr[i] == 1);
}


/* Test worker-single/vector-single mode.  */

void t22()
{
  int arr[32], i;

  for (i = 0; i < 32; i++)
    arr[i] = 0;

  #pragma acc parallel copy(arr) \
		       num_gangs(8) num_workers(8) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      {
	#pragma acc atomic
	arr[j]++;
      }
  }

  for (i = 0; i < 32; i++)
    assert (arr[i] == 1);
}


/* Test condition in worker-single/vector-single mode.  */

void t23()
{
  int arr[32], i;

  for (i = 0; i < 32; i++)
    arr[i] = i;

  #pragma acc parallel copy(arr) \
		       num_gangs(8) num_workers(8) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      if ((arr[j] % 2) != 0)
	arr[j]++;
      else
	arr[j] += 2;
  }

  for (i = 0; i < 32; i++)
    assert (arr[i] == ((i % 2) != 0) ? i + 1 : i + 2);
}


/* Test switch in worker-single/vector-single mode.  */

void t24()
{
  int arr[32], i;

  for (i = 0; i < 32; i++)
    arr[i] = i;

  #pragma acc parallel copy(arr) \
		       num_gangs(8) num_workers(8) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      switch (arr[j] % 5)
	{
	case 0: arr[j] += 1; break;
	case 1: arr[j] += 2; break;
	case 2: arr[j] += 3; break;
	case 3: arr[j] += 4; break;
	case 4: arr[j] += 5; break;
	default: arr[j] += 99;
	}
  }

  for (i = 0; i < 32; i++)
    assert (arr[i] == i + (i % 5) + 1);
}


/* Test worker-single/vector-partitioned mode.  */

void t25()
{
  int arr[32 * 32], i;

  for (i = 0; i < 32 * 32; i++)
    arr[i] = i;

  #pragma acc parallel copy(arr) \
		       num_gangs(8) num_workers(8) vector_length(32)
  {
    int j;
    #pragma acc loop gang
    for (j = 0; j < 32; j++)
      {
	int k;
	#pragma acc loop vector
	for (k = 0; k < 32; k++)
	  {
	    #pragma acc atomic
	    arr[j * 32 + k]++;
	  }
      }
  }

  for (i = 0; i < 32 * 32; i++)
    assert (arr[i] == i + 1);
}


/* Test worker-single, vector-partitioned, gang-redundant mode.  */

#define ACTUAL_GANGS 8
void t27()
{
  int n, arr[32], i;
  int ondev;

  for (i = 0; i < 32; i++)
    arr[i] = 0;

  n = 0;

  #pragma acc parallel copy(n, arr) copyout(ondev) \
	  num_gangs(ACTUAL_GANGS) num_workers(8) vector_length(32)
  {
    int j;

    ondev = acc_on_device (acc_device_not_host);

    #pragma acc atomic
    n++;

    #pragma acc loop vector
    for (j = 0; j < 32; j++)
      {
	#pragma acc atomic
	arr[j] += 1;
      }

    #pragma acc atomic
    n++;
  }

  int m = ondev ? ACTUAL_GANGS : 1;
  
  assert (n == m * 2);

  for (i = 0; i < 32; i++)
    assert (arr[i] == m);
}
#undef ACTUAL_GANGS


/* Check if worker-single variables get broadcastd to vectors.  */

#pragma acc routine
float t28_routine ()
{
  return 2.71;
}

#define N 32
void t28()
{
  float threads[N], v1 = 3.14;

  for (int i = 0; i < N; i++)
    threads[i] = -1;

#pragma acc parallel num_gangs (1) vector_length (32) copy (v1)
  {
    float val = t28_routine ();

#pragma acc loop vector
    for (int i = 0; i < N; i++)
      threads[i] = val + v1*i;
  }

  for (int i = 0; i < N; i++)
    assert (fabs (threads[i] - (t28_routine () + v1*i)) < 0.0001);
}
#undef N


int main()
{
  t1();
  t2();
  t4();
  t5();
  t7();
  t8();
  t9();
  t10();
  t11();
  t12();
  t13();
  t16();
  t17();
  t18();
  t19();
  t20();
  t21();
  t22();
  t23();
  t24();
  t25();
  t27();
  t28();

  return 0;
}