def permute_block_size_according_to_layout(block_size, layout):
"""Returns modified block_size such that the fastest coordinate gets the biggest block dimension"""
if not is_integer_sequence(block_size):
return block_size
sorted_block_size = list(sorted(block_size, reverse=True))
while len(sorted_block_size) > len(layout):
sorted_block_size[0] *= sorted_block_size[-1]
sorted_block_size = sorted_block_size[:-1]
result = list(block_size)
for l, bs in zip(reversed(layout), sorted_block_size):
result[l] = bs
return tuple(result[:len(layout)])
def max_threads_per_block(self):
if is_integer_sequence(self._block_size):
return prod(self._block_size)
else:
return None
def __new__(cls, integer, divisor):
if is_integer_sequence((integer, divisor)):
return integer // divisor
else:
return super().__new__(cls, integer, divisor)
def __new__(cls, integer, divisor):
if is_integer_sequence((integer, divisor)):
return (int(integer) // int(divisor)) * divisor
else:
return super().__new__(cls, integer, divisor)
def has_fixed_index_shape(self):
return is_integer_sequence(self.index_shape)