mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

!40184 Code Check and Doc fix 

Merge pull request !40184 from zichun_ye/op_dev
This commit is contained in:
i-robot 2022-08-10 07:57:40 +00:00 committed by Gitee
parent 1f6eab5596 2def22dd5d
commit c6d3b7b568
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
7 changed files with 83 additions and 77 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
docs/api/api_python/ops/mindspore.ops.func_space_to_batch_nd.rst
View File

@ -17,10 +17,6 @@ mindspore.ops.space_to_batch_nd
w'_i = (w_i + paddings[i][0] + paddings[i][1])//block\_shape[i]
\end{array}
.. note::
Ascend只支持四维张量的输入。
参数：
- **input_x** (Tensor) - 输入张量，Ascend平台必须为四维。
- **block_size** (list[int], tuple[int], int) - 块形状描述空间维度为分割的个数。如果 `block_size` 为list或者tuple，其长度 `M` 为空间维度的长度。如果 `block_size` 为整数，那么所有空间维度分割的个数均为 `block_size` 。在Ascend后端 `M` 必须为2。
@ -34,6 +30,6 @@ mindspore.ops.space_to_batch_nd
- **TypeError** - 如果 `paddings` 不是 list 或者 tuple。
- **ValueError** - 如果当 `block_size` 为 list 或 tuple， `block_size` 不是一维。
- **ValueError** - 如果 Ascend 平台上 `block_size` 长度不是2。
- **ValueError** - 如果 `paddings` 的形状不是 (2, M), 其中 M 为 `block_size` 的长度。
- **ValueError** - 如果 `paddings` 的形状不是 (M, 2), 其中 M 为 `block_size` 的长度。
- **ValueError** - 如果 `block_size` 的元素不是大于一的整数。
- **ValueError** - 如果 `paddings` 的元素不是非负的整数。

3
mindspore/python/mindspore/nn/probability/distribution/categorical.py
View File

@ -153,8 +153,7 @@ class Categorical(Distribution):
self.less = P.Less()
# when the graph kernel mode is enable
# use Log directly as akg will handle the corner cases
self.log = P.Log() if context.get_context(
"enable_graph_kernel") else log_generic
self.log = P.Log() if context.get_context("enable_graph_kernel") else log_generic
self.log_softmax = P.LogSoftmax()
self.logicor = P.LogicalOr()
self.logicand = P.LogicalAnd()

6
mindspore/python/mindspore/nn/probability/distribution/distribution.py
View File

@ -162,8 +162,7 @@ class Distribution(Cell):
self.default_parameters = []
self.parameter_names = []
# cast value to a tensor if it is not None
value_t = None if value is None else cast_to_tensor(
value, self.parameter_type)
value_t = None if value is None else cast_to_tensor(value, self.parameter_type)
self.default_parameters.append(value_t)
self.parameter_names.append(name)
return value_t
@ -184,8 +183,7 @@ class Distribution(Cell):
if arg is not None:
self.checktensor(arg, name)
else:
arg = default if default is not None else raise_none_error(
name)
arg = default if default is not None else raise_none_error(name)
# broadcast if the number of args > 1
if broadcast_shape is None:

3
mindspore/python/mindspore/nn/probability/distribution/normal.py
View File

@ -170,8 +170,7 @@ class Normal(Distribution):
self.expm1 = P.Expm1()
# when the graph kernel mode is enable
# use Log directly as akg will handle the corner cases
self.log = P.Log() if context.get_context(
"enable_graph_kernel") else log_generic
self.log = P.Log() if context.get_context("enable_graph_kernel") else log_generic
self.erf = P.Erf()
self.squeeze = P.Squeeze(0)
self.cast = P.Cast()

2
mindspore/python/mindspore/ops/function/array_func.py
View File

@ -2729,7 +2729,7 @@ def space_to_batch_nd(input_x, block_size, paddings):
ValueError: If `block_size` is not one dimensional when `block_size` is a list or tuple.
ValueError: If the length of `block_size` is not 2 on Ascend.
ValueError: If the element of `block_size` is not an integer larger than 1.
ValueError: If shape of `paddings` is not (2, M), where M is the length of `block_size`.
ValueError: If shape of `paddings` is not (M, 2), where M is the length of `block_size`.
ValueError: If the element of `paddings` is not an integer larger than 0.
TypeError: If `block_size` is not one of list, tuple, int.
TypeError: If `paddings` is neither list nor tuple.

11
mindspore/python/mindspore/ops/operations/_ms_kernel.py
View File

@ -269,7 +269,7 @@ class VariableUsage(ast.NodeVisitor):
else:
decl, loop, usage = self.status.get(node.id, (None, None, None))
usage.add(type(node.ctx))
if not loop in self.scope_level:
if loop not in self.scope_level:
raise ValueError(
"In the function {} written in the Hybrid DSL, there is "
"a variable used out of the scope it is defined: {}".format(self.func_name, node.id))
@ -412,7 +412,7 @@ class VariableUsage(ast.NodeVisitor):
"In the function {} written in the Hybrid DSL, getattr is only supported for a tensor object, "
"not for the object with type: {}".format(self.func_name, type(node.value)))
if not node.value.id in self.output_tensor + self.temp_tensor + list(self.args_index.keys()):
if node.value.id not in self.output_tensor + self.temp_tensor + list(self.args_index.keys()):
raise ValueError(
"In the function {} written in the Hybrid DSL, getattr is only supported for a tensor variable "
"after its declaration, not for: {}".format(self.func_name, node.value.id))
@ -442,16 +442,17 @@ class VariableUsage(ast.NodeVisitor):
"should be the name of a tensor, but get a {}.".format(self.func_name, type(i)))
symbols = list(i.id for i in node.value.elts)
for sy in symbols:
if not sy in list(self.args_index.keys()) + self.output_tensor:
if sy not in list(self.args_index.keys()) + self.output_tensor:
raise TypeError("In the function {} written in the Hybrid DSL, the element in the return value "
"should be either an input tensor or a tensor allocated by output_tensor, "
"but get name: {}".format(self.func_name, sy))
for sy in self.output_tensor:
if not sy in symbols:
if sy not in symbols:
raise TypeError("In the function {} written in the Hybrid DSL, the tensor is allocated as an output "
"tensor but not in the return value: {}".format(self.func_name, sy))
self.inplace_assign_output = list([idx, self.args_index.get(val, -1)]
for idx, val in enumerate(symbols) if val in self.args_index)
for idx, val in enumerate(symbols)
if val in self.args_index)
def determine_variable_usage(root, func_name):

129
mindspore/python/mindspore/ops/operations/custom_ops.py
View File

@ -122,7 +122,7 @@ def _compile_aot(file):
proc = subprocess.Popen(
cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
(out, _) = proc.communicate()
(out, _) = proc.communicate(timeout=30)
if proc.returncode != 0:
msg = "Compilation error in compiling {}:\n".format(file)
@ -146,6 +146,17 @@ class Custom(ops.PrimitiveWithInfer):
.. warning::
This is an experimental prototype that is subject to change.
.. note::
The supported platforms are determined by the input `func_type`:
- "hybrid": supports ["Ascend", "GPU", "CPU"].
- "akg": supports ["Ascend", "GPU", "CPU"].
- "tbe": supports ["Ascend"].
- "aot": supports ["GPU", "CPU"].
- "pyfunc": supports ["CPU"].
- "julia": supports ["CPU"].
- "aicpu": supports ["Ascend"].
Args:
func (Union[function, str]):
@ -496,6 +507,60 @@ class Custom(ops.PrimitiveWithInfer):
self.add_prim_attr("func_type", self.func_type)
self._update_attr()
def __infer__(self, *args):
if callable(self.out_shape):
infer_shape = self.out_shape(*(x["shape"] for x in args))
else:
infer_shape = self.out_shape
if callable(self.out_dtype):
infer_dtype = self.out_dtype(*(x["dtype"] for x in args))
else:
infer_dtype = self.out_dtype
infer_value = None
# deal with the case of ms script
# enable auto infer function if any infer information is missing
if self._is_ms_kernel and (infer_dtype is None or infer_shape is None):
logger.warning("{}, 'out_shape' or 'out_dtype' is None, infer the output shape and output dtype "
"automatically. There might be some Python RuntimeWarning but it wouldn't influence the "
"result.".format(self.log_prefix))
auto_infer_result = self._auto_infer(*args)
# use automatically inferred shape/dtype if the input infer values are null
infer_shape = auto_infer_result[0] if infer_shape is None else infer_shape
infer_dtype = auto_infer_result[1] if infer_dtype is None else infer_dtype
infer_value = auto_infer_result[2]
# deal with case that the custom op is of type pyfunc with empty output
if self.func_type == "pyfunc":
if infer_shape == ():
logger.warning("{}, 'out_shape' is an empty tuple. Add a placeholder instead. "
"Not recommend to use it as it could be any uninitialized data.".format(self.log_prefix))
infer_shape = (1,)
if infer_dtype == ():
logger.warning("{}, 'out_dtype' is an empty tuple. Add a placeholder instead. "
"Not recommend to use it as it could be any uninitialized data.".format(self.log_prefix))
infer_dtype = mstype.int32
# after all automatic infer information fulfillment, throw error if infer_shape/infer_dtype is still None
if not isinstance(infer_shape, (tuple, list)):
raise TypeError("{}, 'out_shape' must be one of [tuple, list, function], but got {}"
.format(self.log_prefix, type(infer_shape)))
if not isinstance(infer_dtype, (typing.Type, tuple, list)):
raise TypeError("{}, 'out_dtype' must be one of [mindspore.dtype, tuple, list, function], but got {}"
.format(self.log_prefix, type(infer_dtype)))
out = {
"shape": infer_shape,
"dtype": infer_dtype,
"value": infer_value,
}
return out
def get_bprop(self):
return self.bprop
@ -560,7 +625,7 @@ class Custom(ops.PrimitiveWithInfer):
"""Update information of func"""
if callable(self.func):
# For the func_type other then hybrid, get the original function if func is decorated
if "__wrapped__" in self.func.__dict__ and not self.func_type in ["hybrid", "pyfunc"]:
if "__wrapped__" in self.func.__dict__ and self.func_type not in ["hybrid", "pyfunc"]:
self.func = self.func.__dict__["__wrapped__"]
# func name
self.func_name = self.func.__name__
@ -590,7 +655,9 @@ class Custom(ops.PrimitiveWithInfer):
inplace_assign_output = determine_variable_usage(root, self.func_name)
if inplace_assign_output:
self.add_prim_attr("inplace_assign_output",
" ".join((str(j) for i in inplace_assign_output for j in i)))
" ".join((str(j)
for i in inplace_assign_output
for j in i)))
self.add_prim_attr('func_source_str', self.func_source_str)
# unique func name
@ -630,7 +697,7 @@ class Custom(ops.PrimitiveWithInfer):
new_dtype_format.append(i + (DataType.I32_Default,))
reg_info["dtype_format"] = new_dtype_format
for i, item in enumerate(reg_info.get("outputs", [])):
for _, item in enumerate(reg_info.get("outputs", [])):
output_name_list = []
if isinstance(item, dict) and item.get("name"):
output_name_list.append(item.get("name"))
@ -936,57 +1003,3 @@ class Custom(ops.PrimitiveWithInfer):
infer_value = Tensor(fake_output) if enable_infer_value else None
return infer_shape, infer_dtype, infer_value
def __infer__(self, *args):
if callable(self.out_shape):
infer_shape = self.out_shape(*(x["shape"] for x in args))
else:
infer_shape = self.out_shape
if callable(self.out_dtype):
infer_dtype = self.out_dtype(*(x["dtype"] for x in args))
else:
infer_dtype = self.out_dtype
infer_value = None
# deal with the case of ms script
# enable auto infer function if any infer information is missing
if self._is_ms_kernel and (infer_dtype is None or infer_shape is None):
logger.warning("{}, 'out_shape' or 'out_dtype' is None, infer the output shape and output dtype "
"automatically. There might be some Python RuntimeWarning but it wouldn't influence the "
"result.".format(self.log_prefix))
auto_infer_result = self._auto_infer(*args)
# use automatically inferred shape/dtype if the input infer values are null
infer_shape = auto_infer_result[0] if infer_shape is None else infer_shape
infer_dtype = auto_infer_result[1] if infer_dtype is None else infer_dtype
infer_value = auto_infer_result[2]
# deal with case that the custom op is of type pyfunc with empty output
if self.func_type == "pyfunc":
if infer_shape == ():
logger.warning("{}, 'out_shape' is an empty tuple. Add a placeholder instead. "
"Not recommend to use it as it could be any uninitialized data.".format(self.log_prefix))
infer_shape = (1,)
if infer_dtype == ():
logger.warning("{}, 'out_dtype' is an empty tuple. Add a placeholder instead. "
"Not recommend to use it as it could be any uninitialized data.".format(self.log_prefix))
infer_dtype = mstype.int32
# after all automatic infer information fulfillment, throw error if infer_shape/infer_dtype is still None
if not isinstance(infer_shape, (tuple, list)):
raise TypeError("{}, 'out_shape' must be one of [tuple, list, function], but got {}"
.format(self.log_prefix, type(infer_shape)))
if not isinstance(infer_dtype, (typing.Type, tuple, list)):
raise TypeError("{}, 'out_dtype' must be one of [mindspore.dtype, tuple, list, function], but got {}"
.format(self.log_prefix, type(infer_dtype)))
out = {
"shape": infer_shape,
"dtype": infer_dtype,
"value": infer_value,
}
return out