update activation op on kunlun (#29577)

* fix expand && concat/transpose to new api

* update xpu_header

* update activation op on kunlun

* update activation op on kunlun

* update activation op on kunlun

* update activation op on kunlun

* update activation op on kunlun

* add nearest_interp on kunlun

* update error message

cmake/external/xpu.cmake

@@ -4,7 +4,7 @@ endif()
 
 INCLUDE(ExternalProject)
 SET(XPU_PROJECT                 "extern_xpu")
-SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2020_12_11.tar.gz" CACHE STRING "" FORCE)
+SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2020_12_15.tar.gz" CACHE STRING "" FORCE)
 SET(XPU_SOURCE_DIR              "${THIRD_PARTY_PATH}/xpu")
 SET(XPU_DOWNLOAD_DIR            "${XPU_SOURCE_DIR}/src/${XPU_PROJECT}")
 SET(XPU_INSTALL_DIR             "${THIRD_PARTY_PATH}/install/xpu")

paddle/fluid/operators/interpolate_op_xpu.cc

@@ -229,9 +229,7 @@ class InterpolateGradXPUKernel : public framework::OpKernel<T> {
     int trans_mode = (align_corners) ? (0) : ((align_mode == 0) ? (1) : (2));
 
     if (nearest) {
-      PADDLE_ENFORCE_EQ((data_layout == DataLayout::kNCHW), true,
-                        platform::errors::InvalidArgument(
-                            "XPU nearest is only support NCHW"));
+      trans_mode = (align_corners) ? (0) : (2);
     }
 
     r = xpu::interpolate2d_grad<T>(dev_ctx.x_context(), output_grad->data<T>(),
@@ -252,7 +250,10 @@ class InterpolateGradXPUKernel : public framework::OpKernel<T> {
 namespace ops = paddle::operators;
 
 REGISTER_OP_XPU_KERNEL(bilinear_interp, ops::InterpolateXPUKernel<float>);
+REGISTER_OP_XPU_KERNEL(nearest_interp, ops::InterpolateXPUKernel<float>);
 
 REGISTER_OP_XPU_KERNEL(bilinear_interp_grad,
                        ops::InterpolateGradXPUKernel<float>);
+REGISTER_OP_XPU_KERNEL(nearest_interp_grad,
+                       ops::InterpolateGradXPUKernel<float>);
 #endif

python/paddle/fluid/tests/unittests/xpu/test_activation_op_xpu.py

@@ -73,8 +73,7 @@ class TestXPUSigmoid(TestXPUActivation):
     def test_check_grad(self):
         if paddle.is_compiled_with_xpu():
             place = paddle.XPUPlace(0)
-            self.check_grad_with_place(
-                place, ['X'], 'Out', max_relative_error=0.01)
+            self.check_grad_with_place(place, ['X'], 'Out')
 
 
 @unittest.skipIf(not paddle.is_compiled_with_xpu(),
@@ -90,6 +89,11 @@ class TestXPUTanh(TestXPUActivation):
         self.inputs = {'X': OpTest.np_dtype_to_fluid_dtype(x)}
         self.outputs = {'Out': out}
 
+    def test_check_grad(self):
+        if paddle.is_compiled_with_xpu():
+            place = paddle.XPUPlace(0)
+            self.check_grad_with_place(place, ['X'], 'Out')
+
 
 @unittest.skipIf(not paddle.is_compiled_with_xpu(),
                  "core is not compiled with XPU")
@@ -105,6 +109,11 @@ class TestXPUSqrt(TestXPUActivation):
         self.inputs = {'X': OpTest.np_dtype_to_fluid_dtype(x)}
         self.outputs = {'Out': out}
 
+    def test_check_grad(self):
+        if paddle.is_compiled_with_xpu():
+            place = paddle.XPUPlace(0)
+            self.check_grad_with_place(place, ['X'], 'Out')
+
 
 @unittest.skipIf(not paddle.is_compiled_with_xpu(),
                  "core is not compiled with XPU")
@@ -142,6 +151,11 @@ class TestXPURelu(TestXPUActivation):
         self.inputs = {'X': x}
         self.outputs = {'Out': out}
 
+    def test_check_grad(self):
+        if paddle.is_compiled_with_xpu():
+            place = paddle.XPUPlace(0)
+            self.check_grad_with_place(place, ['X'], 'Out')
+
 
 @unittest.skipIf(not paddle.is_compiled_with_xpu(),
                  "core is not compiled with XPU")
@@ -157,6 +171,11 @@ class TestXPUGelu(TestXPUActivation):
         self.outputs = {'Out': out}
         self.attrs = {"approximate": approximate, 'use_xpu': True}
 
+    def test_check_grad(self):
+        if paddle.is_compiled_with_xpu():
+            place = paddle.XPUPlace(0)
+            self.check_grad_with_place(place, ['X'], 'Out')
+
 
 def gelu(x, approximate):
     if approximate:
@@ -223,6 +242,11 @@ class TestXPUSquare(TestXPUActivation):
         self.inputs = {'X': OpTest.np_dtype_to_fluid_dtype(x)}
         self.outputs = {'Out': out}
 
+    def test_check_grad(self):
+        if paddle.is_compiled_with_xpu():
+            place = paddle.XPUPlace(0)
+            self.check_grad_with_place(place, ['X'], 'Out')
+
 
 @unittest.skipIf(not paddle.is_compiled_with_xpu(),
                  "core is not compiled with XPU")
@@ -239,5 +263,36 @@ class TestXPUPow(TestXPUActivation):
         self.outputs = {'Out': out}
 
 
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestXPULeakyRelu(TestXPUActivation):
+    def setUp(self):
+        self.op_type = "leaky_relu"
+        self.init_dtype()
+        x = np.random.uniform(-1, 1, [11, 17]).astype(self.dtype)
+        alpha = np.random.uniform(
+            0,
+            1, )
+        out = leaky_relu(x, alpha)
+
+        self.inputs = {'X': x}
+        self.outputs = {'Out': out}
+        self.attrs = {'use_xpu': True, 'alpha': alpha}
+
+    def test_check_grad(self):
+        if paddle.is_compiled_with_xpu():
+            place = paddle.XPUPlace(0)
+            self.check_grad_with_place(place, ['X'], 'Out')
+
+
+def leaky_relu(x, alpha):
+    if (alpha < 1):
+        y_ref = np.maximum(x, alpha * x)
+    else:
+        y_ref = np.minimum(x, alpha * x)
+    return y_ref.astype(x.dtype)
+
+
 if __name__ == "__main__":
+    paddle.enable_static()
     unittest.main()