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

!1563 Fixing some tiny faults about Pylint in my code(ops) 

Merge pull request !1563 from liuwenhao/master
This commit is contained in:
mindspore-ci-bot 2020-05-29 15:04:51 +08:00 committed by Gitee
parent 0b3da2c787 f87c70a9ab
commit d3dbb10b6d
25 changed files with 150 additions and 168 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
mindspore/ops/_op_impl/_custom_op/batch_matmul_impl.py
View File

@ -164,9 +164,10 @@ def CusBatchMatMul(input_x1, input_x2, output, transpose_a=False, transpose_b=Tr
matmul_hybrid_f_t_local_UB = tik_instance.Tensor(dtype, [64],
name="matmul_hybrid_f_t_local_UB",
scope=tik.scope_ubuf)
matmul_hybrid_f_t_local_UB_dst_tmp = tik_instance.Tensor(dtype, [64],
name="matmul_hybrid_f_t_local_UB_dst_tmp",
scope=tik.scope_ubuf)
matmul_hybrid_f_t_local_UB_dst_tmp = tik_instance.Tensor(
dtype, [64],
name="matmul_hybrid_f_t_local_UB_dst_tmp",
scope=tik.scope_ubuf)
tik_instance.vector_dup(64, matmul_hybrid_f_t_local_UB, 0, 1, 1, 8)
tik_instance.data_move(input_2_local_UB,
input2[(block_idx // 6) * 16384 + thread_idx2 * 8192], 0, 1,

33
mindspore/ops/_op_impl/_custom_op/matmul_cube_dense_left_impl.py
View File

@ -127,7 +127,7 @@ def _shape_check(shape_a, shape_b, shape_bias, src_dtype, trans_a, trans_b):
if n_shape % cce.BLOCK_IN != 0 and n_shape != 1:
raise RuntimeError("input shape N should be 1 or multiple of %d" % cce.BLOCK_IN)
if len(shape_bias) != 0:
if shape_bias:
if len(shape_bias) == 1:
if is_gevm or is_gemv:
if shape_bias[0] != m_shape * n_shape:
@ -189,7 +189,7 @@ def check_supported(input_x1, input_x2, bias=None, output_y={}, trans_a=False, t
util.check_shape_size(shape_b, SHAPE_SIZE_LIMIT)
try:
trans_a_f = bool(1 - trans_a)
if src_dtype == "float32" or src_dtype == "int32":
if src_dtype in ("float32", "int32"):
if len(shape_a) != 2 and len(shape_b) != 2:
return False
if trans_b:
@ -239,6 +239,7 @@ def check_supported(input_x1, input_x2, bias=None, output_y={}, trans_a=False, t
return False
except RuntimeError as e:
print(e)
return False
return True
@ -385,7 +386,7 @@ def CusMatMulCubeDenseLeft(input_x1, input_x2, bias=None, output_y={}, trans_a=F
tensor_b = tvm.placeholder(shape_b_temp, name='tensor_b',
dtype=src_dtype)
if len(shape_bias) > 0:
if shape_bias:
tensor_bias = tvm.placeholder(shape_bias, name='tensor_bias',
dtype=dst_dtype)
@ -449,20 +450,20 @@ def CusMatMulCubeDenseLeft(input_x1, input_x2, bias=None, output_y={}, trans_a=F
resMatmul_local_UB, 0, 16, 224 // 2, 0, 56 * 16 * 2 // 2)
tik_instance.BuildCCE(kernel_name=kernel_name, inputs=[input_x1, input_x2], outputs=[resMatmul])
return tik_instance
else:
print("come into tbe, shape is error!")
result = te.lang.cce.matmul(tensor_a, tensor_b, trans_a, trans_b, format_a=format_a,
format_b=format_b, dst_dtype=dst_dtype, tensor_bias=tensor_bias)
with tvm.target.cce():
schedule = generic.auto_schedule(result)
print("come into tbe, shape is error!")
result = te.lang.cce.matmul(tensor_a, tensor_b, trans_a, trans_b, format_a=format_a,
format_b=format_b, dst_dtype=dst_dtype, tensor_bias=tensor_bias)
tensor_list = [tensor_a, tensor_b, result]
if len(shape_bias) > 0:
tensor_list = [tensor_a, tensor_b, tensor_bias, result]
with tvm.target.cce():
schedule = generic.auto_schedule(result)
config = {"print_ir": False,
"name": kernel_name,
"tensor_list": tensor_list}
tensor_list = [tensor_a, tensor_b, result]
if shape_bias:
tensor_list = [tensor_a, tensor_b, tensor_bias, result]
te.lang.cce.cce_build_code(schedule, config)
config = {"print_ir": False,
"name": kernel_name,
"tensor_list": tensor_list}
te.lang.cce.cce_build_code(schedule, config)

14
mindspore/ops/_op_impl/_custom_op/matmul_cube_fracz_left_cast_impl.py
View File

@ -124,7 +124,7 @@ src_dtype: str
if n_shape % cce.BLOCK_IN != 0 and n_shape != 1:
raise RuntimeError("input shape N should be 1 or multiple of %d" % cce.BLOCK_IN)
if len(shape_bias):
if shape_bias:
if len(shape_bias) == 1:
if is_gevm or is_gemv:
if shape_bias[0] != m_shape * n_shape:
@ -144,11 +144,10 @@ def _get_bias(shape_bias):
bias_length = shape_bias[0]
if bias_length % 16 == 0:
return shape_bias
else:
bias_length = (bias_length // 16) * 16 + 16
shape_bias = []
shape_bias.append(bias_length)
return shape_bias
bias_length = (bias_length // 16) * 16 + 16
shape_bias = []
shape_bias.append(bias_length)
return shape_bias
def _get_input_shape(shape_x):
@ -184,7 +183,7 @@ def check_supported(input_x1, input_x2, bias=None, output_y={}, trans_a=False, t
util.check_shape_size(shape_b, SHAPE_SIZE_LIMIT)
try:
trans_a_f = bool(1 - trans_a)
if src_dtype == "float32" or src_dtype == "int32":
if src_dtype in ("floate32", "int32"):
if len(shape_a) != 2 and len(shape_b) != 2:
return False
if trans_b:
@ -234,6 +233,7 @@ def check_supported(input_x1, input_x2, bias=None, output_y={}, trans_a=False, t
return False
except RuntimeError as e:
print(e)
return False
return True

4
mindspore/ops/_op_impl/_custom_op/matmul_cube_fracz_right_mul_impl.py
View File

@ -80,8 +80,8 @@ def CusMatMulCubeFraczRightMul(input_x1, input_x2, input_x3, bias=None, output_y
((32, 128, 16, 16), 'float16', (32, 32, 16, 16), 'float16', (1,), 'float32'),
((64, 32, 16, 16), 'float16', (64, 64, 16, 16), 'float16', (1,), 'float32'),
((16, 64, 16, 16), 'float16', (16, 16, 16, 16), 'float16', (1,), 'float32')]
input_shape = (
tuple(input_x1_shape), input_x1_dtype, tuple(input_x2_shape), input_x2_dtype, tuple(input_x3_shape), input_x3_dtype)
input_shape = (tuple(input_x1_shape), input_x1_dtype, tuple(input_x2_shape),
input_x2_dtype, tuple(input_x3_shape), input_x3_dtype)
if input_shape not in Supported:
raise RuntimeError("input_shape %s is not supported" % str(input_shape))

20
mindspore/ops/_op_impl/_custom_op/matmul_cube_impl.py
View File

@ -129,7 +129,7 @@ def _shape_check(shape_a, shape_b, shape_bias, src_dtype, trans_a, trans_b):
if n_shape % cce.BLOCK_IN != 0 and n_shape != 1:
raise RuntimeError("input shape N should be 1 or multiple of %d" % cce.BLOCK_IN)
if len(shape_bias):
if shape_bias:
if len(shape_bias) == 1:
if is_gevm or is_gemv:
if shape_bias[0] != m_shape * n_shape:
@ -149,11 +149,10 @@ def _get_bias(shape_bias):
bias_length = shape_bias[0]
if bias_length % 16 == 0:
return shape_bias
else:
bias_length = (bias_length // 16) * 16 + 16
shape_bias = []
shape_bias.append(bias_length)
return shape_bias
bias_length = (bias_length // 16) * 16 + 16
shape_bias = []
shape_bias.append(bias_length)
return shape_bias
def _get_input_shape(shape_x):
@ -189,7 +188,7 @@ def check_supported(input_x1, input_x2, bias=None, output_y={}, trans_a=False, t
util.check_shape_size(shape_b, SHAPE_SIZE_LIMIT)
try:
trans_a_f = bool(1 - trans_a)
if src_dtype == "float32" or src_dtype == "int32":
if src_dtype in ("float32", "int32"):
if len(shape_a) != 2 and len(shape_b) != 2:
return False
if trans_b:
@ -239,6 +238,7 @@ def check_supported(input_x1, input_x2, bias=None, output_y={}, trans_a=False, t
return False
except RuntimeError as e:
print(e)
return False
return True
@ -314,7 +314,7 @@ def CusMatMulCube(input_x1, input_x2, bias=None, output_y={}, trans_a=False, tra
src_dtype = input_x1.get("dtype").lower()
dst_dtype = output_y.get("dtype").lower()
if src_dtype == "float32" or src_dtype == "int32":
if src_dtype in ("float32", "int32"):
matmul_vector_cce(shape_a, shape_b, src_dtype, trans_a, trans_b, shape_bias, kernel_name)
return
_shape_check(shape_a, shape_b, shape_bias, src_dtype, trans_a, trans_b)
@ -377,7 +377,7 @@ def CusMatMulCube(input_x1, input_x2, bias=None, output_y={}, trans_a=False, tra
tensor_b = tvm.placeholder(shape_b_temp, name='tensor_b',
dtype=src_dtype)
if len(shape_bias) > 0:
if shape_bias:
tensor_bias = tvm.placeholder(shape_bias, name='tensor_bias',
dtype=dst_dtype)
result = te.lang.cce.matmul(tensor_a, tensor_b, trans_a, trans_b, format_a=format_a,
@ -387,7 +387,7 @@ def CusMatMulCube(input_x1, input_x2, bias=None, output_y={}, trans_a=False, tra
schedule = generic.auto_schedule(result)
tensor_list = [tensor_a, tensor_b, result]
if len(shape_bias) > 0:
if shape_bias:
tensor_list = [tensor_a, tensor_b, tensor_bias, result]
config = {"print_ir": False,

17
tests/st/ops/ascend/test_ops_infer.py
View File

@ -16,17 +16,10 @@
import functools
import numpy as np
import mindspore.nn as nn
import mindspore.context as context
import mindspore.common.dtype as mstype
from mindspore import Tensor, Parameter
from mindspore.common.initializer import initializer
from mindspore.ops import Primitive
from mindspore.ops import composite as C
from mindspore import Tensor
from mindspore.ops import operations as P
from mindspore.ops import functional as F
from mindspore.ops import prim_attr_register, PrimitiveWithInfer
from mindspore.ops.primitive import constexpr
from mindspore import context
context.set_context(mode=context.GRAPH_MODE, save_graphs=True)
@ -38,7 +31,7 @@ def test_cast_op_attr():
self.cast = P.Cast()
def construct(self, x, t):
return self.cast(x, t)
class CastTypeTest(nn.Cell):
def __init__(self, net):
super(CastTypeTest, self).__init__()
@ -54,9 +47,9 @@ def test_cast_op_attr():
t5 = cast_net(z, mstype.float16)
return (t1, t2, t3, t4, t5)
net = CastTypeTest(CastNet())
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.int32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1918]).astype(np.int32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.int32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.int32))
out = net(t1, t2, t3)
assert out[0].asnumpy().dtype == np.float32
assert out[1].asnumpy().dtype == np.int32

10
tests/st/ops/ascend/test_tbe_ops/test_mul.py
View File

@ -33,13 +33,13 @@ class Net(nn.Cell):
return self.mul(x1, x2)
x1 = np.random.randn(3, 4).astype(np.float32)
x2 = np.random.randn(3, 4).astype(np.float32)
arr_x1 = np.random.randn(3, 4).astype(np.float32)
arr_x2 = np.random.randn(3, 4).astype(np.float32)
def test_net():
mul = Net()
output = mul(Tensor(x1), Tensor(x2))
print(x1)
print(x2)
output = mul(Tensor(arr_x1), Tensor(arr_x2))
print(arr_x1)
print(arr_x2)
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_npu_clear_float_status.py
View File

@ -33,11 +33,11 @@ class Net(nn.Cell):
return self.npu_clear_float_status(x1)
x1 = np.random.randn(8).astype(np.float32)
arr_x1 = np.random.randn(8).astype(np.float32)
def test_net():
npu_clear_float_status = Net()
output = npu_clear_float_status(Tensor(x1))
print(x1)
output = npu_clear_float_status(Tensor(arr_x1))
print(arr_x1)
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_npu_get_float_status.py
View File

@ -33,11 +33,11 @@ class Net(nn.Cell):
return self.npu_get_float_status(x1)
x1 = np.random.randn(8).astype(np.float32)
arr_x1 = np.random.randn(8).astype(np.float32)
def test_net():
npu_get_float_status = Net()
output = npu_get_float_status(Tensor(x1))
print(x1)
output = npu_get_float_status(Tensor(arr_x1))
print(arr_x1)
print(output.asnumpy())

4
tests/st/ops/ascend/test_tbe_ops/test_pad.py
View File

@ -34,11 +34,11 @@ class Net(nn.Cell):
return x
x = np.random.random(size=(2, 2)).astype(np.float32)
arr_x = np.random.random(size=(2, 2)).astype(np.float32)
def test_net():
pad = Net()
output = pad(Tensor(x))
output = pad(Tensor(arr_x))
print("=================output====================")
print(output.asnumpy())

10
tests/st/ops/ascend/test_tbe_ops/test_realdiv.py
View File

@ -33,13 +33,13 @@ class Net(nn.Cell):
return self.realdiv(x1, x2)
x1 = np.random.randn(3, 4).astype(np.float32)
x2 = np.random.randn(3, 4).astype(np.float32)
arr_x1 = np.random.randn(3, 4).astype(np.float32)
arr_x2 = np.random.randn(3, 4).astype(np.float32)
def test_net():
realdiv = Net()
output = realdiv(Tensor(x1), Tensor(x2))
print(x1)
print(x2)
output = realdiv(Tensor(arr_x1), Tensor(arr_x2))
print(arr_x1)
print(arr_x2)
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_reciprocal.py
View File

@ -33,11 +33,11 @@ class Net(nn.Cell):
return self.reciprocal(x1)
x1 = np.random.randn(3, 4).astype(np.float32)
arr_x1 = np.random.randn(3, 4).astype(np.float32)
def test_net():
reciprocal = Net()
output = reciprocal(Tensor(x1))
print(x1)
output = reciprocal(Tensor(arr_x1))
print(arr_x1)
print(output.asnumpy())

10
tests/st/ops/ascend/test_tbe_ops/test_scatter_nd.py
View File

@ -31,13 +31,13 @@ class Net(nn.Cell):
return self.scatternd(indices, update, (3, 3))
indices = np.array([[0, 1], [1, 1]]).astype(np.int32)
update = np.array([3.2, 1.1]).astype(np.float32)
arr_indices = np.array([[0, 1], [1, 1]]).astype(np.int32)
arr_update = np.array([3.2, 1.1]).astype(np.float32)
def test_net():
scatternd = Net()
print(indices)
print(update)
output = scatternd(Tensor(indices), Tensor(update))
print(arr_indices)
print(arr_update)
output = scatternd(Tensor(arr_indices), Tensor(arr_update))
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_softmax.py
View File

@ -31,11 +31,11 @@ class Net(nn.Cell):
return self.Softmax(x)
x = np.array([[5, 1]]).astype(np.float32)
arr_x = np.array([[5, 1]]).astype(np.float32)
def test_net():
softmax = Net()
output = softmax(Tensor(x))
print(x)
output = softmax(Tensor(arr_x))
print(arr_x)
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_split.py
View File

@ -31,13 +31,13 @@ class Net(nn.Cell):
return self.split(x)
x = np.random.randn(2, 4).astype(np.float32)
arr_x = np.random.randn(2, 4).astype(np.float32)
def test_net():
split = Net()
output = split(Tensor(x))
output = split(Tensor(arr_x))
print("====input========")
print(x)
print(arr_x)
print("====output=======")
print(output)

6
tests/st/ops/ascend/test_tbe_ops/test_sqrt.py
View File

@ -31,11 +31,11 @@ class Net(nn.Cell):
return self.sqrt(x)
x = np.array([1.0, 4.0, 9.0]).astype(np.float32)
arr_x = np.array([1.0, 4.0, 9.0]).astype(np.float32)
def test_net():
sqrt = Net()
output = sqrt(Tensor(x))
print(x)
output = sqrt(Tensor(arr_x))
print(arr_x)
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_square.py
View File

@ -31,11 +31,11 @@ class Net(nn.Cell):
return self.square(x)
x = np.array([1.0, 4.0, 9.0]).astype(np.float32)
arr_x = np.array([1.0, 4.0, 9.0]).astype(np.float32)
def test_net():
square = Net()
output = square(Tensor(x))
print(x)
output = square(Tensor(arr_x))
print(arr_x)
print(output.asnumpy())

10
tests/st/ops/ascend/test_tbe_ops/test_sub.py
View File

@ -31,13 +31,13 @@ class Net(nn.Cell):
return self.sub(x, y)
x = np.random.randn(1, 3, 3, 4).astype(np.float32)
y = np.random.randn(1, 3, 3, 4).astype(np.float32)
arr_x = np.random.randn(1, 3, 3, 4).astype(np.float32)
arr_y = np.random.randn(1, 3, 3, 4).astype(np.float32)
def test_net():
sub = Net()
output = sub(Tensor(x), Tensor(y))
print(x)
print(y)
output = sub(Tensor(arr_x), Tensor(arr_y))
print(arr_x)
print(arr_y)
print(output.asnumpy())

6
tests/st/ops/ascend/test_tbe_ops/test_tile.py
View File

@ -31,11 +31,11 @@ class Net(nn.Cell):
return self.tile(x, (1, 4))
x = np.array([[0], [1], [2], [3]]).astype(np.int32)
arr_x = np.array([[0], [1], [2], [3]]).astype(np.int32)
def test_net():
tile = Net()
print(x)
output = tile(Tensor(x))
print(arr_x)
output = tile(Tensor(arr_x))
print(output.asnumpy())

2
tests/st/ops/cpu/test_addn_op.py
View File

@ -68,7 +68,7 @@ def test_net_3Input():
addn = Net3I()
output = addn(Tensor(x, mstype.float32), Tensor(y, mstype.float32), Tensor(z, mstype.float32))
print("output:\n", output)
expect_result = [[0., 3., 6.],
expect_result = [[0., 3., 6.],
[9., 12., 15]]
assert (output.asnumpy() == expect_result).all()

2
tests/st/ops/cpu/test_conv2d_backprop_input_op.py
View File

@ -66,7 +66,7 @@ class Net5(nn.Cell):
def test_conv2d_backprop_input():
conv2d_input = Net5()
output = conv2d_input()
print("================================")
print("================================")
# expect output:
# [[[[ -5, -4, 5, 12, 0, -8]
# [-15, -6, 17, 17, -2, -11]

12
tests/ut/python/ops/test_control_ops.py
View File

@ -20,7 +20,6 @@ import mindspore as ms
from mindspore import Tensor
from mindspore import context
from mindspore import nn
from mindspore.common.parameter import Parameter, ParameterTuple
from mindspore.ops import composite as C
from mindspore.ops import functional as F
from mindspore.ops import operations as P
@ -447,11 +446,14 @@ def test_index_to_switch_layer():
def test_control_depend_check():
with pytest.raises(TypeError) as e:
depend = P.ControlDepend(0.0)
P.ControlDepend(0.0)
print(e)
with pytest.raises(ValueError) as e:
depend = P.ControlDepend(2)
P.ControlDepend(2)
print(e)
with pytest.raises(TypeError) as e:
depend = P.ControlDepend((2,))
P.ControlDepend((2,))
print(e)
def test_if_nested_compile():
@ -497,7 +499,7 @@ def test_if_inside_for():
c1 = Tensor(1, dtype=ms.int32)
c2 = Tensor(1, dtype=ms.int32)
net = Net()
out = net(c1, c2)
net(c1, c2)
def test_while_in_while():

14
tests/ut/python/ops/test_nn_ops.py
View File

@ -31,7 +31,6 @@ from ....mindspore_test_framework.pipeline.forward.compile_forward \
import pipeline_for_compile_forward_ge_graph_for_case_by_case_config
from ....mindspore_test_framework.pipeline.forward.verify_exception \
import pipeline_for_verify_exception_for_case_by_case_config
from mindspore import context
context.set_context(mode=context.GRAPH_MODE, save_graphs=True)
def conv3x3(in_channels, out_channels, stride=1, padding=1):
@ -382,17 +381,18 @@ def test_conv2d_same_primitive():
class Conv2DSameNet(nn.Cell):
def __init__(self):
super(Conv2DSameNet, self).__init__()
self.conv1 = nn.Conv2d(16, 64, (1, 41), (1,4), "same", 0, 1, has_bias=True)
self.conv2 = nn.Conv2d(16, 64, (1, 41), (1,4), "same", 0, 1, has_bias=True)
self.conv1 = nn.Conv2d(16, 64, (1, 41), (1, 4), "same", 0, 1, has_bias=True)
self.conv2 = nn.Conv2d(16, 64, (1, 41), (1, 4), "same", 0, 1, has_bias=True)
def construct(self, x, y):
r1 = self.conv1(x)
r2 = self.conv2(y)
return (r1, r2)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
net = Conv2DSameNet()
out = net(t1, t2)
net(t1, t2)
class ComparisonNet(nn.Cell):
def __init__(self):
""" ComparisonNet definition """

99
tests/ut/python/ops/test_ops_attr_infer.py
View File

@ -13,30 +13,14 @@
# limitations under the License.
# ============================================================================
""" test nn ops """
import functools
import numpy as np
import mindspore
import mindspore.nn as nn
import mindspore.context as context
import mindspore.common.dtype as mstype
from mindspore import Tensor, Parameter
from mindspore.common.initializer import initializer
from mindspore.ops import Primitive
from mindspore.ops import composite as C
from mindspore.ops import operations as P
from mindspore import Tensor
from mindspore.ops import functional as F
from mindspore.ops import prim_attr_register, PrimitiveWithInfer
from mindspore.ops.primitive import constexpr
from ..ut_filter import non_graph_engine
from ....mindspore_test_framework.mindspore_test import mindspore_test
from ....mindspore_test_framework.pipeline.forward.compile_forward \
import pipeline_for_compile_forward_ge_graph_for_case_by_case_config
from ....mindspore_test_framework.pipeline.forward.verify_exception \
import pipeline_for_verify_exception_for_case_by_case_config
from mindspore import context
context.set_context(mode=context.GRAPH_MODE, save_graphs=True)
class FakeOp(PrimitiveWithInfer):
@ -57,16 +41,16 @@ def test_conv2d_same_primitive():
class Conv2DSameNet(nn.Cell):
def __init__(self):
super(Conv2DSameNet, self).__init__()
self.conv1 = nn.Conv2d(16, 64, (1, 41), (1,4), "same", 0, 1, has_bias=True)
self.conv2 = nn.Conv2d(16, 64, (1, 41), (1,4), "same", 0, 1, has_bias=True)
self.conv1 = nn.Conv2d(16, 64, (1, 41), (1, 4), "same", 0, 1, has_bias=True)
self.conv2 = nn.Conv2d(16, 64, (1, 41), (1, 4), "same", 0, 1, has_bias=True)
def construct(self, x, y):
r1 = self.conv1(x)
r2 = self.conv2(y)
return (r1, r2)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
net = Conv2DSameNet()
out = net(t1, t2)
net(t1, t2)
# test cell as high order argument
# The graph with free variables used as argument is not supported yet
@ -87,10 +71,10 @@ def Xtest_conv2d_op_with_arg():
a = self.opnet(conv_op, x)
b = self.opnet(conv_op, y)
return (a, b)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
net = OpsNet(Conv2dNet())
out = net(t1, t2)
net(t1, t2)
def test_conv2d_op_with_arg():
@ -115,11 +99,10 @@ def test_conv2d_op_with_arg():
a = self.opnet(op, x, y)
b = self.opnet(op, y, x)
return (a, b)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
net = OpsNet(OpNet())
out = net(t1, t2)
net(t1, t2)
def test_conv2d_op_with_arg_same_input():
@ -144,10 +127,10 @@ def test_conv2d_op_with_arg_same_input():
a = self.opnet(op, x, x)
b = self.opnet(op, y, x)
return (a, b)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
net = OpsNet(OpNet())
out = net(t1, t2)
net(t1, t2)
# test op with partial
def test_op_as_partial():
@ -160,11 +143,11 @@ def test_op_as_partial():
a = partial_op(y)
b = partial_op(z)
return a, b
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1234]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1234]).astype(np.float32))
net = OpAsPartial()
out = net(t1, t2, t3)
net(t1, t2, t3)
# test op with partial
def test_op_as_partial_inside():
@ -182,13 +165,14 @@ def test_op_as_partial_inside():
super(OuterNet, self).__init__()
self.net = OpAsPartial()
def construct(self, x, y, z):
a,b = self.net(x, y, z)
a, b = self.net(x, y, z)
return a, b
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1234]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1234]).astype(np.float32))
net = OuterNet()
out = net(t1, t2, t3)
net(t1, t2, t3)
# test op with partial case 2
def test_op_as_partial_independent():
@ -202,11 +186,12 @@ def test_op_as_partial_independent():
partial_op2 = F.partial(self.op, x)
b = partial_op2(z)
return a, b
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1234]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1234]).astype(np.float32))
net = OpAsPartial()
out = net(t1, t2, t3)
net(t1, t2, t3)
def test_nest_partial():
class NestPartial(nn.Cell):
@ -221,11 +206,11 @@ def test_nest_partial():
partial_op4 = F.partial(partial_op3, x)
b = partial_op4(z)
return a, b
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1234]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1234]).astype(np.float32))
net = NestPartial()
out = net(t1, t2, t3)
net(t1, t2, t3)
# high order argument
# op and op args as network arguments
@ -245,11 +230,11 @@ def test_op_with_arg_as_input():
a = self.opnet(op, x, z)
b = self.opnet(op, x, y)
return (a, b)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1234]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1234]).astype(np.float32))
net = OpsNet(WithOpArgNet())
out = net(t1, t2, t3)
net(t1, t2, t3)
# The partial application used as argument is not supported yet
# because of the limit of inference specialize system
@ -269,8 +254,8 @@ def Xtest_partial_as_arg():
a = self.partial_net(partial_op, z)
b = self.partial_net(partial_op, y)
return (a, b)
t1 = Tensor(np.ones([1,16,1,1918]).astype(np.float32))
t2 = Tensor(np.ones([1,16,1,3840]).astype(np.float32))
t3 = Tensor(np.ones([1,16,1,1234]).astype(np.float32))
t1 = Tensor(np.ones([1, 16, 1, 1918]).astype(np.float32))
t2 = Tensor(np.ones([1, 16, 1, 3840]).astype(np.float32))
t3 = Tensor(np.ones([1, 16, 1, 1234]).astype(np.float32))
net = OpsNet(PartialArgNet())
out = net(t1, t2, t3)
net(t1, t2, t3)

2
tests/ut/python/pynative_mode/test_framstruct.py
View File

@ -982,7 +982,7 @@ def test_bprop_with_wrong_output_shape():
@bprop_getters.register(BpropWithWrongOutputShape)
def get_bprop_with_wrong_output_shape(self):
"""Generate bprop for BpropWithWrongOutputShape"""
ones = Tensor(np.ones([2, ]).astype(np.int32))
ones = Tensor(np.ones([2,]).astype(np.int32))
def bprop(x, out, dout):
return (ones,)