Add elementwise_mul orig2prim and support p_norm when p=1

python/paddle/autograd/primrules.py

@@ -111,6 +111,7 @@ def linear_jvp(op, *args, **kwargs):
 elementwise_sub
 scale
 assign
+elementwise_mul
 
 These original ops are partially supported:
 
@@ -122,6 +123,24 @@ def linear_jvp(op, *args, **kwargs):
 """
 
 
+@REGISTER_ORIG2PRIM('elementwise_mul')
+def elementwise_mul_orig2prim(op, x, y):
+    if x.shape != y.shape:
+        y = broadcast(y, shape=x.shape)
+    if op.attr('Scale_x') - 1.0 > 1e-5:
+        tmp = fill_const(shape=x.shape, dtype=x.dtype, value=op.attr('Scale_x'))
+        x = mul(x, tmp)
+    if op.attr('Scale_y') - 1.0 > 1e-5:
+        tmp = fill_const(shape=y.shape, dtype=y.dtype, value=op.attr('Scale_y'))
+        y = mul(y, tmp)
+    z = mul(x, y)
+    if op.attr('Scale_out') - 1.0 > 1e-5:
+        tmp = fill_const(
+            shape=z.shape, dtype=z.dtype, value=op.attr('Scale_out'))
+        z = mul(z, tmp)
+    return z
+
+
 @REGISTER_ORIG2PRIM('matmul_v2')
 def matmul_v2_orig2prim(op, x, y):
     def trans(shape):
@@ -222,13 +241,17 @@ def num_el(shape):
             n = n * s
         return n
 
-    assert op.attr(
-        'porder') - 2.0 < 1e-5, 'Only support lower l2 norm currently'
     assert op.attr(
         'asvector'), 'Only support lower pnorm when asvector=True currently'
     if len(x.shape) > 1:
         x = reshape(x, shape=[num_el(x.shape)])
-    return sqrt(reduce(mul(x, x), axis=[0]))
+
+    if op.attr('porder') - 2.0 < 1e-5:
+        return sqrt(reduce(mul(x, x), axis=[0]))
+    elif op.attr('porder') - 1.0 < 1e-5:
+        return reduce(sqrt(mul(x, x)), axis=[0])
+    else:
+        raise RuntimeError('Only support lower l2/l1 norm currently')
 
 
 @REGISTER_ORIG2PRIM('index_select')

python/paddle/fluid/tests/unittests/test_orig2prig_rules.py

@@ -68,6 +68,25 @@ def test_op(self):
                 self.assertEqual(prim_out[k].shape, v.shape)
 
 
+class TestElementWiseMulOrig2Prim(TestElementWiseAddOrig2Prim):
+    def init_data(self):
+        self.op_type = 'elementwise_mul'
+        X = paddle.static.data(name='X', shape=[8, 8], dtype='float')
+        Y = paddle.static.data(name='Y', shape=[8, 8], dtype='float')
+
+        self.input = {'X': X, 'Y': Y}
+        self.output = {
+            'Out':
+            self.layer_help.create_variable_for_type_inference(dtype=X.dtype)
+        }
+        self.attrs = {}
+
+        self.orig2prim_args = (X, Y)
+        self.all_ops = ['elementwise_mul', 'mul_p']
+        # { prim_op_output_index: orig_op_output_var }
+        self.out_map = {0: self.output['Out']}
+
+
 class TestMatmulV2Orig2Prim(TestElementWiseAddOrig2Prim):
     def init_data(self):
         self.op_type = 'matmul_v2'
@@ -201,7 +220,27 @@ def init_data(self):
         self.out_map = {0: self.output['Out']}
 
 
-class TestPNormOrig2Prim(TestElementWiseAddOrig2Prim):
+class TestPNormOrig2Prim1(TestElementWiseAddOrig2Prim):
+    def init_data(self):
+        self.op_type = 'p_norm'
+        X = paddle.static.data(name='X', shape=[5, 6], dtype='int64')
+
+        self.input = {'X': X, }
+        self.output = {
+            'Out':
+            self.layer_help.create_variable_for_type_inference(dtype=X.dtype)
+        }
+        self.attrs = {
+            'porder': 1,
+            'asvector': True,
+        }
+
+        self.orig2prim_args = (X, )
+        self.all_ops = ['p_norm', 'reshape_p', 'sqrt_p', 'reduce_p', 'mul_p']
+        self.out_map = {0: self.output['Out']}
+
+
+class TestPNormOrig2Prim2(TestElementWiseAddOrig2Prim):
     def init_data(self):
         self.op_type = 'p_norm'
         X = paddle.static.data(name='X', shape=[5, 6], dtype='int64')