Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion devito/passes/iet/definitions.py
Original file line number Diff line number Diff line change
Expand Up @@ -601,7 +601,7 @@ def _alloc_local_array_on_high_bw_mem(self, site, obj, storage):
doalloc = self.langbb['device-alloc']
dofree = self.langbb['device-free']

nbytes = SizeOf(obj._C_typedata)*obj.size
nbytes = SizeOf(obj._C_typedata)*as_long(obj.size)

zeroing, efuncs = self._zero_init(obj, storage)
allocs = [doalloc(nbytes, deviceid, retobj=obj), *zeroing]
Expand Down
24 changes: 24 additions & 0 deletions tests/test_dse.py
Original file line number Diff line number Diff line change
Expand Up @@ -2976,6 +2976,30 @@ def test_subdims_wo_block_temps(self):

op.apply()

@pytest.mark.parametrize('lang', ['C', 'CXX'])
def test_temp_array_allocation_uses_long(self, lang):
"""
Ensure that operands are cast to long when calculating the size of a
temporary array to prevent int32 wraparound.
"""
grid = Grid(shape=(31, 17, 25))

a = Function(name='a', grid=grid)
f = TimeFunction(name='f', grid=grid)
g = TimeFunction(name='g', grid=grid)

eq0 = Eq(f.forward, g + sin(a).dx)
eq1 = Eq(g.forward, f + sin(a).dx)

opt = ('advanced', {'cire-minmem': True, 'index-mode': 'int64'})
with switchconfig(language=lang):
op = Operator([eq0, eq1], opt=opt)

cast = '(long)' if lang == 'C' else 'static_cast<long>'

# 3 casts in memalign, 3 in memset
assert str(op).count(cast) == 6


class TestIsoAcoustic:

Expand Down
3 changes: 2 additions & 1 deletion tests/test_gpu_openmp.py
Original file line number Diff line number Diff line change
Expand Up @@ -217,7 +217,8 @@ def test_array_rw(self):
assert len(op.body.allocs) == 2
assert str(op.body.allocs[0]) ==\
('float * r0_vec = (float *)'
'omp_target_alloc(x_size*y_size*z_size*sizeof(float),'
'omp_target_alloc('
'sizeof(float)*(long)z_size*(long)y_size*(long)x_size,'
'omp_get_default_device());')
assert str(op.body.allocs[1]) ==\
'init0(x_M,x_m,y_M,y_m,z_M,z_m,r0_vec,y_size,z_size);'
Expand Down
Loading