"""utils for operator"""
from mindspore.common.tensor import Tensor
from ..._checkparam import Validator as validator
from ..._checkparam import Rel
from ...common import dtype as mstype
from ..primitive import constexpr
def get_broadcast_shape(x_shape, y_shape, prim_name, shape_type=""):
"""
Doing broadcast between tensor x and tensor y.
Args:
x_shape (list): The shape of tensor x.
y_shape (list): The shape of tensor y.
prim_name (str): Primitive name.
Returns:
List, the shape that broadcast between tensor x and tensor y.
Raises:
ValueError: If tensor x and tensor y are not equal and couldn't broadcast.
Examples:
>>> x_shape = [1, 2, 3]
>>> y_shape = [1, 2]
>>> broadcast_shape = get_broadcast_shape(x_shape, y_shape)
"""
if x_shape == y_shape:
return x_shape
x_len = len(x_shape)
y_len = len(y_shape)
length = x_len if x_len < y_len else y_len
broadcast_shape_back = []
for i in range(-length, 0):
if x_shape[i] == 1:
broadcast_shape_back.append(y_shape[i])
elif y_shape[i] == 1:
broadcast_shape_back.append(x_shape[i])
elif x_shape[i] == y_shape[i]:
broadcast_shape_back.append(x_shape[i])
elif x_shape[i] == -1 or y_shape[i] == -1:
broadcast_shape_back.append(-1)
else:
if shape_type == "min_shape":
broadcast_shape_back.append(max(x_shape[i], y_shape[i]))
elif shape_type == "max_shape":
broadcast_shape_back.append(min(x_shape[i], y_shape[i]))
else:
raise ValueError(f"For '{prim_name}', x_shape and y_shape are supposed to broadcast, "
f"where broadcast means that "
f"x_shape[i] = 1 or -1 or y_shape[i] = 1 or -1 or x_shape[i] = y_shape[i], "
f"but now x_shape and y_shape can not broadcast, "
f"got i: {i}, x_shape: {x_shape}, y_shape: {y_shape}.")
broadcast_shape_front = y_shape[0: y_len - length] if length == x_len else x_shape[0: x_len - length]
broadcast_shape = list(broadcast_shape_front) + broadcast_shape_back
return broadcast_shape
def get_concat_offset(x_shp, x_type, axis, prim_name):
"""for concat and concatoffset check args and compute offset"""
validator.check_value_type("shape", x_shp, [tuple, list], prim_name)
validator.check_positive_int(len(x_shp), "input_x rank", prim_name)
validator.check_subclass("shape0", x_type[0], mstype.tensor, prim_name)
validator.check_positive_int(len(x_shp[0]), "len of x_shp[0]", prim_name)
rank_base = len(x_shp[0])
validator.check_int_range(axis, -rank_base - 1, rank_base, Rel.INC_BOTH, 'axis', prim_name)
if axis < 0:
axis = axis + rank_base
all_shp = x_shp[0][axis]
offset = [0]
for i in range(1, len(x_shp)):
v = x_shp[i]
validator.check('len of x_shp[%d]' % i, len(v), 'len of x_shp[0]', len(x_shp[0]), Rel.EQ, prim_name)
validator.check('x_type[%d]' % i, x_type[i], 'x_type[0]', x_type[0], Rel.EQ, prim_name)
for j in range(rank_base):
if j != axis and v[j] != x_shp[0][j]:
raise ValueError(f"The shape of the two input elements of the Concat operator do not match:"
f"shape[0] = {x_shp[0]} and shape[1] = {x_shp[1]}.")
offset.append(all_shp)
if all_shp == -1 or v[axis] == -1:
all_shp = -1
else:
all_shp += v[axis]
return offset, all_shp, axis
@constexpr
def range_op(start, limit, delta, dtype):
"""helper function to get tensor in specified range."""
output_tensor = Tensor(list(range(start, limit, delta)), dtype)
return output_tensor
@constexpr
def get_1d_shape(in_shape):
"""helper function to get 1d shape."""
out_shape = 1
for i in in_shape:
out_shape *= i
return (out_shape,)
@constexpr
def generate_shape_index(out_shape, indices_shape, axis):
out_rank = len(out_shape)
ind_rank = len(indices_shape)
if axis < 0:
axis += out_rank - ind_rank + 1
perm_part1 = tuple(range(axis, axis + ind_rank))
index = tuple(range(out_rank))
perm = perm_part1 + index[:axis] + index[axis + ind_rank:]
return perm