def test_global_offsets_consts_array(self):
k = KernelCode('''
double Y[s][n];
double F[s][n];
double A[s][s];
double y[n];
for (int j = 0; j < n; ++j)
{
Y[0][j] += A[0][0] * F[0][j];
Y[0][j] += A[0][1] * F[1][j];
Y[1][j] += A[1][0] * F[0][j];
Y[1][j] += A[1][1] * F[1][j];
Y[0][j] = Y[0][j] + y[j];
Y[1][j] = Y[1][j] + y[j];
}''',
machine=None)
k.set_constant('n', 100000000)
k.set_constant('s', 2)
offsets_warmup = k.compile_global_offsets(iteration=range(0, 10000),
spacing=0)
for l, s in offsets_warmup:
self.assertEqual(len(l), 5, msg="Number of load offsets")
self.assertEqual(len(s), 2, msg="Number of store offsets")
def test_global_offsets_2d(self):
k = KernelCode(self.twod_code, machine=None)
k.set_constant('N', 10)
k.set_constant('M', 20)
sizes = k.array_sizes(in_bytes=True, subs_consts=True)
offsets = k.compile_global_offsets(iteration=0, spacing=0)
read_offsets, write_offsets = list(offsets)[0]
# read access to a[j][i-1], a[j][i+1], a[j-1][i], a[j+1][i]
self.assertCountEqual([(1 * 10 + 0) * 8, (1 * 10 + 2) * 8,
(0 * 10 + 1) * 8, (2 * 10 + 1) * 8],
read_offsets)
# write access to b[i][j]
self.assertCountEqual([sizes['a'] + (1 * 10 + 1) * 8], write_offsets)
def test_global_offsets_2d(self):
k = KernelCode(self.twod_code, machine=None)
k.set_constant('N', 10)
k.set_constant('M', 20)
sizes = k.array_sizes(in_bytes=True, subs_consts=True)
offsets = k.compile_global_offsets(iteration=0, spacing=0)
read_offsets, write_offsets = list(offsets)[0]
# read access to a[j][i-1], a[j][i+1], a[j-1][i], a[j+1][i]
self.assertCountEqual(
[(1 * 10 + 0) * 8, (1 * 10 + 2) * 8, (0 * 10 + 1) * 8, (2 * 10 + 1) * 8],
read_offsets)
# write access to b[i][j]
self.assertCountEqual(
[sizes['a'] + (1 * 10 + 1) * 8],
write_offsets)
def test_global_offsets_variable_small_array(self):
k = KernelCode('''
double Y[s][n];
double y[n];
for (int l = 0; l < s; l++)
for (int j = 0; j < n; j++)
Y[l][j] = Y[l][j] + y[j];''', machine=None)
k.set_constant('n', 100000000)
k.set_constant('s', 2)
offsets_warmup = k.compile_global_offsets(iteration=range(0, 10000), spacing=0)
for l, s in offsets_warmup:
self.assertEqual(len(l), 2, msg="Number of load offsets")
self.assertEqual(len(s), 1, msg="Number of store offsets")
def test_global_offsets_consts_array(self):
k = KernelCode('''
double Y[s][n];
double F[s][n];
double A[s][s];
double y[n];
for (int j = 0; j < n; ++j)
{
Y[0][j] += A[0][0] * F[0][j];
Y[0][j] += A[0][1] * F[1][j];
Y[1][j] += A[1][0] * F[0][j];
Y[1][j] += A[1][1] * F[1][j];
Y[0][j] = Y[0][j] + y[j];
Y[1][j] = Y[1][j] + y[j];
}''', machine=None)
k.set_constant('n', 100000000)
k.set_constant('s', 2)
offsets_warmup = k.compile_global_offsets(iteration=range(0,10000), spacing=0)
for l,s in offsets_warmup:
self.assertEqual(len(l), 5, msg="Number of load offsets")
self.assertEqual(len(s), 2, msg="Number of store offsets")
