Fix bias_add gradient (apache#4516)

* Fix bias_add gradient

A change caused collapse_sum_like to reject implicit dimension
broadcasting for bias_add gradient, so switch to explicit sum reduction
on the non-bias axis dimensions.

* Lint fix

python/tvm/relay/op/_tensor_grad.py

@@ -379,9 +379,9 @@ def log_softmax_grad(orig, grad):
 @register_gradient("nn.bias_add")
 def bias_add_grad(orig, grad):
     """Returns gradient of bias_add"""
-    data, bias = orig.args
+    data = orig.args[0]
     return [collapse_sum_like(grad, data),
-            collapse_sum_like(grad, bias)]
+            _sum(grad, orig.attrs.axis, keepdims=False, exclude=True)]
 
 
 @register_gradient("nn.dense")

tests/python/relay/test_op_grad_level1.py

@@ -110,12 +110,19 @@ def test_log_softmax_grad():
     check_grad(fwd_func, scale=1)
 
 
-def test_bias_add_grad():
-    data = relay.var("data", relay.TensorType((1, 16), "float32"))
-    bias = relay.var("bias", relay.TensorType((16,), "float32"))
-    fwd_func = relay.Function([data, bias], relay.nn.bias_add(data, bias))
+def verify_bias_add(d_shape, b_shape, axis=1):
+    data = relay.var("data", relay.TensorType(d_shape, "float32"))
+    bias = relay.var("bias", relay.TensorType(b_shape, "float32"))
+    fwd_func = relay.Function([data, bias], relay.nn.bias_add(data, bias, axis=axis))
     check_grad(fwd_func)
 
 
+def test_bias_add_grad():
+    verify_bias_add((1, 16), (16,))
+    verify_bias_add((1, 8, 2, 2), (8,))
+    verify_bias_add((1, 2, 2, 8), (8,), 3)
+    verify_bias_add((4, 8), (8,))
+
+
 if __name__ == "__main__":
     pytest.main([__file__])