fix slice_assign_p (#47324)

* fix slice_assign_p jvp and transpose

* modify code style

* modify test_jvp_and_transpose for slice_assign_p

* modify code style

* add unit test

python/paddle/fluid/tests/unittests/autograd/test_jvp_and_transpose.py

@@ -883,7 +883,7 @@ class TestSliceSelectPJVPAndTranspose(TestAddPJVPAndTranspose):
         ]
 
 
-class TestSliceAssignPJVPAndTranspose(TestAddPJVPAndTranspose):
+class TestSliceAssignPJVPAndTranspose1(TestAddPJVPAndTranspose):
     def init_data(self):
         # Set prim op
         self.op_type = 'slice_assign_p'
@@ -909,23 +909,110 @@ class TestSliceAssignPJVPAndTranspose(TestAddPJVPAndTranspose):
         self.jvp_out_shape_map = {0: self.prim_output['Z']}
 
         # Set transpose
-        check_dot = lambda v: v is X or v is Y
+        check_dot = lambda v: v is X
         Z_BAR = paddle.static.data(name='Z_BAR', shape=[3, 20], dtype='float64')
         self.transpose_args = (check_dot, Z_BAR)
-        self.transpose_out_shape_map = {0: X, 1: Y}
+        self.transpose_out_shape_map = {0: X}
 
         self.all_ops = [
             # prim op:
             'slice_assign_p',
             # jvp op:
             'slice_assign_p',
+            "slice_assign_p",
+            "add_p",
+            "fill_constant_p",
+            "fill_constant_p",
             # transpose op:
             'slice_assign_p',
+            'fill_constant_p',
+        ]
+
+
+class TestSliceAssignPJVPAndTranspose2(TestAddPJVPAndTranspose):
+    def init_data(self):
+        # Set prim op
+        self.op_type = 'slice_assign_p'
+        X = paddle.static.data(name='X', shape=[3, 20], dtype='float64')
+        Y = paddle.static.data(name='Y', shape=[3, 5], dtype='float64')
+        self.prim_input = {'X': X, 'Y': Y}
+        self.prim_output = {
+            'Z': self.layer_help.create_variable_for_type_inference(
+                dtype=X.dtype
+            )
+        }
+        self.prim_attrs = {
+            'axis': [1],
+            'starts': [0],
+            'ends': [10],
+            'strides': [2],
+        }
+
+        # Set JVP
+        Y_DOT = paddle.static.data(name='Y_DOT', shape=[3, 5], dtype='float64')
+        self.jvp_args = (None, Y_DOT)
+        self.jvp_out_shape_map = {0: self.prim_output['Z']}
+
+        # Set transpose
+        check_dot = lambda v: v is Y
+        Z_BAR = paddle.static.data(name='Z_BAR', shape=[3, 20], dtype='float64')
+        self.transpose_args = (check_dot, Z_BAR)
+        self.transpose_out_shape_map = {1: Y}
+
+        self.all_ops = [
+            # prim op:
+            'slice_assign_p',
+            # jvp op:
+            'slice_assign_p',
+            "fill_constant_p",
+            # transpose op:
             'slice_select_p',
             'fill_constant_p',
         ]
 
 
+class TestSliceAssignPJVPAndTranspose3(TestAddPJVPAndTranspose):
+    def init_data(self):
+        # Set prim op
+        self.op_type = 'slice_assign_p'
+        X = paddle.static.data(name='X', shape=[3, 20], dtype='float64')
+        Y = paddle.static.data(name='Y', shape=[3, 5], dtype='float64')
+        self.prim_input = {'X': X, 'Y': Y}
+        self.prim_output = {
+            'Z': self.layer_help.create_variable_for_type_inference(
+                dtype=X.dtype
+            )
+        }
+        self.prim_attrs = {
+            'axis': [1],
+            'starts': [0],
+            'ends': [10],
+            'strides': [2],
+        }
+
+        # Set JVP
+        X_DOT = paddle.static.data(name='X_DOT', shape=[3, 20], dtype='float64')
+        self.jvp_args = (X_DOT, None)
+        self.jvp_out_shape_map = {0: self.prim_output['Z']}
+
+        # Set transpose
+        check_dot = lambda v: v is X
+        Z_BAR = paddle.static.data(name='Z_BAR', shape=[3, 20], dtype='float64')
+        self.transpose_args = (check_dot, Z_BAR)
+        self.transpose_out_shape_map = {0: X}
+
+        self.all_ops = [
+            # prim op:
+            'slice_assign_p',
+            # jvp op:
+            'slice_assign_p',
+            "fill_constant_p",
+            # transpose op:
+            'slice_assign_p',
+            'fill_constant_p',
+        ]
+
+
 class TestGatherPJVPAndTranspose(TestAddPJVPAndTranspose):
     def init_data(self):
         # Set prim op

python/paddle/fluid/tests/unittests/autograd/test_primapi.py

@@ -100,7 +100,7 @@ class TestFowardApi(unittest.TestCase):
         actual = actual()
         self.assertEqual(type(actual), type(expected))
         for i, j in zip(actual, expected):
-            np.testing.assert_allclose(i, j, atol=1e-3, rtol=1e-3)
+            np.testing.assert_allclose(i, j, rtol=1e-6)
 
 
 @utils.place(config.DEVICES)

python/paddle/incubate/autograd/primrules.py

@@ -1026,17 +1026,53 @@ def slice_select_jvp(op, x_dot):
 
 @REGISTER_JVP('slice_assign_p')
 def slice_assign_jvp(op, x_dot, y_dot):
-    if x_dot is None:
-        assert y_dot is None, 'y_dot must be None.'
-        return None
-    else:
-        assert y_dot is not None, 'y_dot should not be None.'
+    x, y = op_position_inputs(op)
+    assert (
+        x_dot is not None or y_dot is not None
+    ), "x_dot and y_dot can't be None at the same time. "
     axis = op.attr('axis')
     starts = op.attr('starts')
     ends = op.attr('ends')
     strides = op.attr('strides')
+    if x_dot is None:
         return linear_jvp(
-        op, x_dot, y_dot, axis=axis, starts=starts, ends=ends, strides=strides
+            op,
+            fill_const(value=0.0, shape=x.shape, dtype=x.dtype),
+            y_dot,
+            axis=axis,
+            starts=starts,
+            ends=ends,
+            strides=strides,
+        )
+    elif y_dot is None:
+        return linear_jvp(
+            op,
+            x_dot,
+            fill_const(value=0.0, shape=y.shape, dtype=y.dtype),
+            axis=axis,
+            starts=starts,
+            ends=ends,
+            strides=strides,
+        )
+    return add(
+        linear_jvp(
+            op,
+            fill_const(value=0.0, shape=x.shape, dtype=x.dtype),
+            y_dot,
+            axis=axis,
+            starts=starts,
+            ends=ends,
+            strides=strides,
+        ),
+        linear_jvp(
+            op,
+            x_dot,
+            fill_const(value=0.0, shape=y.shape, dtype=y.dtype),
+            axis=axis,
+            starts=starts,
+            ends=ends,
+            strides=strides,
+        ),
     )
 
 
@@ -1311,8 +1347,8 @@ def slice_select_transpose(op, check_dot, y_bar):
 @REGISTER_TRANSPOSE('slice_assign_p')
 def slice_assign_transpose(op, check_dot, z_bar):
     x, y = op_position_inputs(op)
-    assert check_dot(x) and check_dot(y), (
-        f'(check_dot(x) and check_dot(y)) must be True, '
+    assert check_dot(x) ^ check_dot(y), (
+        f'(check_dot(x) ^ check_dot(y)) must be True, '
         f'but check_dot(x)={check_dot(x)} and check_dot(y)={check_dot(y)}.'
     )
     zeros = fill_const(value=0.0, shape=y.shape, dtype=y.dtype)
@@ -1320,13 +1356,21 @@ def slice_assign_transpose(op, check_dot, z_bar):
     starts = op.attr('starts')
     ends = op.attr('ends')
     strides = op.attr('strides')
-    x_bar = slice_assign(
-        z_bar, zeros, axis=axis, starts=starts, ends=ends, strides=strides
+    if check_dot(x):
+        return (
+            slice_assign(
+                z_bar,
+                zeros,
+                axis=axis,
+                starts=starts,
+                ends=ends,
+                strides=strides,
+            ),
+            None,
         )
-    y_bar = slice_select(
+    return None, slice_select(
         z_bar, axis=axis, starts=starts, ends=ends, strides=strides
     )
-    return x_bar, y_bar
 
 
 @REGISTER_TRANSPOSE('gather_p')