update activation op on kunlun (#29577) (#29717)

* 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/activation_op_xpu.cc

@@ -54,55 +54,27 @@ class XPUActivationGradKernel
 };
 
 template <typename DeviceContext, typename T>
-void xpu_activation_forward(const framework::ExecutionContext &ctx,
-                            xpu::Activation_t type) {
+void xpu_activation_forward(
+    const framework::ExecutionContext &ctx,
+    std::function<int(xpu::Context *, const T *, T *, int)> func) {
   const auto *x = ctx.Input<Tensor>("X");
   auto *y = ctx.Output<Tensor>("Out");
   const T *x_data = x->data<T>();
   T *y_data = y->mutable_data<T>(ctx.GetPlace());
-  int r = 0;
-  auto xpu_context = ctx.device_context<DeviceContext>().x_context();
 
-  switch (type.type) {
-    case xpu::Activation_t::HARD_SWISH: {
-      float threshold = ctx.Attr<float>("threshold");
-      float scale = ctx.Attr<float>("scale");
-      float offset = ctx.Attr<float>("offset");
-      PADDLE_ENFORCE_EQ(threshold, 6.0f,
-                        platform::errors::External(
-                            "Not support threshold [%f] in XPU", threshold));
-      PADDLE_ENFORCE_EQ(
-          scale, 6.0f,
-          platform::errors::External("Not support scale [%f] in XPU", scale));
+  auto xpu_context = ctx.device_context<DeviceContext>().x_context();
+  int r = func(xpu_context, x_data, y_data, x->numel());
   PADDLE_ENFORCE_EQ(
-          offset, 3.0f,
-          platform::errors::External("Not support offset [%f] in XPU", offset));
-
-      r = xpu::hard_swish(xpu_context, reinterpret_cast<const float *>(x_data),
-                          reinterpret_cast<float *>(y_data), x->numel());
-      break;
-    }
-    case xpu::Activation_t::ACT_POW: {
-      type.pow_factor = ctx.Attr<float>("factor");
-    }
-    default: {
-      r = xpu::activation_forward(xpu_context, type, x->numel(),
-                                  reinterpret_cast<const float *>(x_data),
-                                  reinterpret_cast<float *>(y_data));
-      break;
-    }
-  }
-
-  PADDLE_ENFORCE_EQ(r, XPU_SUCCESS,
-                    platform::errors::External(
-                        "XPU API return wrong value[%d], please check whether "
-                        "Baidu Kunlun Card is properly installed.",
-                        r));
+      r, xpu::Error_t::SUCCESS,
+      platform::errors::External("XPU activation op return wrong value[%d %s].",
+                                 r, XPUAPIErrorMsg[r]));
 }
 
 template <typename DeviceContext, typename T>
 void xpu_activation_backward(const framework::ExecutionContext &ctx,
-                             xpu::Activation_t type) {
+                             std::function<int(xpu::Context *, const T *,
+                                               const T *, const T *, T *, int)>
+                                 func) {
   /* TODO: relu tanh sigmoid are inplace */
   const auto *x = ctx.Input<Tensor>("X");
   auto *y = ctx.Input<Tensor>("Out");
@@ -115,99 +87,248 @@ void xpu_activation_backward(const framework::ExecutionContext &ctx,
   if (y != nullptr) y_data = y->data<T>();
   if (dOut != nullptr) y_grad = dOut->data<T>();
   T *x_grad = dX->mutable_data<T>(ctx.GetPlace());
-  int r = 0;
   auto xpu_context = ctx.device_context<DeviceContext>().x_context();
 
-  switch (type.type) {
-    case xpu::Activation_t::HARD_SWISH: {
-      float threshold = ctx.Attr<float>("threshold");
-      float scale = ctx.Attr<float>("scale");
-      float offset = ctx.Attr<float>("offset");
-      PADDLE_ENFORCE_EQ(threshold, 6.0f,
+  int r = func(xpu_context, x_data, y_data, y_grad, x_grad, dX->numel());
+  PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
                     platform::errors::External(
-                            "Not support threshold [%f] in XPU", threshold));
-      PADDLE_ENFORCE_EQ(
-          scale, 6.0f,
-          platform::errors::External("Not support scale [%f] in XPU", scale));
-      PADDLE_ENFORCE_EQ(
-          offset, 3.0f,
-          platform::errors::External("Not support offset [%f] in XPU", offset));
-      r = xpu::hard_swish_grad(xpu_context,
-                               reinterpret_cast<const float *>(x_data),
-                               reinterpret_cast<const float *>(y_data),
-                               reinterpret_cast<const float *>(y_grad),
-                               reinterpret_cast<float *>(x_grad), dX->numel());
-      break;
-    }
-    default: {
-      r = xpu::activation_backward(xpu_context, type, dX->numel(),
-                                   reinterpret_cast<const float *>(x_data),
-                                   reinterpret_cast<const float *>(y_data),
-                                   reinterpret_cast<const float *>(y_grad),
-                                   reinterpret_cast<float *>(x_grad));
-      break;
-    }
-  }
-
-  PADDLE_ENFORCE_EQ(r, XPU_SUCCESS,
-                    platform::errors::External(
-                        "XPU API return wrong value[%d], please check whether "
-                        "Baidu Kunlun Card is properly installed.",
-                        r));
+                        "XPU activation grad op return wrong value[%d %s].", r,
+                        XPUAPIErrorMsg[r]));
 }
 
-template <typename T, xpu::Activation_t::act_enum algorithm>
-struct XPUActivationFunc : public BaseActivationFunctor<T> {
+template <typename T>
+struct XPUReluFunctor : public BaseActivationFunctor<T> {
   void operator()(const framework::ExecutionContext &ctx) const {
     xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(ctx,
-                                                                  algorithm);
+                                                                  xpu::relu<T>);
   }
 };
 
-template <typename T, xpu::Activation_t::act_enum algorithm>
-struct XPUActivationGradFunc : public BaseActivationFunctor<T> {
+template <typename T>
+struct XPUSigmoidFunctor : public BaseActivationFunctor<T> {
   void operator()(const framework::ExecutionContext &ctx) const {
-    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(ctx,
-                                                                   algorithm);
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::sigmoid<T>);
   }
 };
 
 template <typename T>
-using XPUReluFunctor = XPUActivationFunc<T, xpu::Activation_t::RELU>;
+struct XPUTanhFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(ctx,
+                                                                  xpu::tanh<T>);
+  }
+};
+
 template <typename T>
-using XPUSigmoidFunctor = XPUActivationFunc<T, xpu::Activation_t::SIGMOID>;
+struct XPUGeluFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(ctx,
+                                                                  xpu::gelu<T>);
+  }
+};
+
 template <typename T>
-using XPUTanhFunctor = XPUActivationFunc<T, xpu::Activation_t::TANH>;
+struct XPULogFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(ctx,
+                                                                  xpu::log<T>);
+  }
+};
+
 template <typename T>
-using XPUGeluFunctor = XPUActivationFunc<T, xpu::Activation_t::GELU>;
+struct XPUSquareFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::square<T>);
+  }
+};
+
 template <typename T>
-using XPULogFunctor = XPUActivationFunc<T, xpu::Activation_t::LOG>;
+struct XPUSqrtFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(ctx,
+                                                                  xpu::sqrt<T>);
+  }
+};
+
 template <typename T>
-using XPUSquareFunctor = XPUActivationFunc<T, xpu::Activation_t::SQUARE>;
+struct XPUAbsFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(ctx,
+                                                                  xpu::abs<T>);
+  }
+};
+
 template <typename T>
-using XPUHardSwishFunctor = XPUActivationFunc<T, xpu::Activation_t::HARD_SWISH>;
+struct XPUPowFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    const auto *x = ctx.Input<Tensor>("X");
+    auto *y = ctx.Output<Tensor>("Out");
+    auto pow_factor = ctx.Attr<float>("factor");
+    const T *x_data = x->data<T>();
+    T *y_data = y->mutable_data<T>(ctx.GetPlace());
+    T *factor_data = nullptr;
+
+    auto xpu_context =
+        ctx.device_context<paddle::platform::XPUDeviceContext>().x_context();
+    PADDLE_ENFORCE_EQ(xpu_malloc(reinterpret_cast<void **>(&factor_data),
+                                 x->numel() * sizeof(T)),
+                      XPU_SUCCESS, platform::errors::ResourceExhausted(
+                                       "XPU has no enough memory"));
+    int r = xpu::constant<T>(xpu_context, factor_data, x->numel(), pow_factor);
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External("XPU constant op return"
+                                   " wrong value[%d %s] in pow op.",
+                                   r, XPUAPIErrorMsg[r]));
+    r = xpu::pow(xpu_context, x_data, factor_data, y_data, x->numel());
+    PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
+                      platform::errors::External("XPU pow op return"
+                                                 " wrong value[%d %s].",
+                                                 r, XPUAPIErrorMsg[r]));
+    if (xpu_context->xpu_stream != nullptr) {
+      xpu_wait(xpu_context->xpu_stream);
+    }
+    xpu_free(factor_data);
+  }
+};
+
 template <typename T>
-using XPUSuareGradFunctor = XPUActivationGradFunc<T, xpu::Activation_t::SQUARE>;
+struct XPUHardSwishFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    float threshold = ctx.Attr<float>("threshold");
+    float scale = ctx.Attr<float>("scale");
+    float offset = ctx.Attr<float>("offset");
+    PADDLE_ENFORCE_EQ(threshold, 6.0f,
+                      platform::errors::External(
+                          "Not support threshold [%f] in XPU", threshold));
+    PADDLE_ENFORCE_EQ(scale, 6.0f, platform::errors::External(
+                                       "Not support scale [%f] in XPU", scale));
+    PADDLE_ENFORCE_EQ(
+        offset, 3.0f,
+        platform::errors::External("Not support offset [%f] in XPU", offset));
+    xpu_activation_forward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::hard_swish<T>);
+  }
+};
+
 template <typename T>
-using XPUReluGradFunctor = XPUActivationGradFunc<T, xpu::Activation_t::RELU>;
+struct XPUReluGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::relu_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUSigmoidGradFunctor =
-    XPUActivationGradFunc<T, xpu::Activation_t::SIGMOID>;
+struct XPUTanhGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::tanh_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUTanhGradFunctor = XPUActivationGradFunc<T, xpu::Activation_t::TANH>;
+struct XPUSigmoidGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::sigmoid_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUGeluGradFunctor = XPUActivationGradFunc<T, xpu::Activation_t::GELU>;
+struct XPUGeluGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::gelu_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUSqrtFunctor = XPUActivationFunc<T, xpu::Activation_t::SQRT>;
+struct XPUSqrtGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::sqrt_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUSqrtGradFunctor = XPUActivationGradFunc<T, xpu::Activation_t::SQRT>;
+struct XPUSquareGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::square_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUHardSwishGradFunctor =
-    XPUActivationGradFunc<T, xpu::Activation_t::HARD_SWISH>;
+struct XPUHardSwishGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    float threshold = ctx.Attr<float>("threshold");
+    float scale = ctx.Attr<float>("scale");
+    float offset = ctx.Attr<float>("offset");
+    PADDLE_ENFORCE_EQ(threshold, 6.0f,
+                      platform::errors::External(
+                          "Not support threshold [%f] in XPU", threshold));
+    PADDLE_ENFORCE_EQ(scale, 6.0f, platform::errors::External(
+                                       "Not support scale [%f] in XPU", scale));
+    PADDLE_ENFORCE_EQ(
+        offset, 3.0f,
+        platform::errors::External("Not support offset [%f] in XPU", offset));
+    xpu_activation_backward<paddle::platform::XPUDeviceContext, T>(
+        ctx, xpu::hard_swish_grad<T>);
+  }
+};
+
 template <typename T>
-using XPUACTPowFunctor = XPUActivationFunc<T, xpu::Activation_t::ACT_POW>;
+struct XPULeakyReluFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    const auto *x = ctx.Input<Tensor>("X");
+    auto *y = ctx.Output<Tensor>("Out");
+    float alpha = ctx.Attr<float>("alpha");
+    const T *x_data = x->data<T>();
+    T *y_data = y->mutable_data<T>(ctx.GetPlace());
+
+    auto xpu_context =
+        ctx.device_context<paddle::platform::XPUDeviceContext>().x_context();
+    int r = xpu::leaky_relu(xpu_context, x_data, y_data, x->numel(), alpha);
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External("XPU leaky_relu return wrong value[%d %s].",
+                                   r, XPUAPIErrorMsg[r]));
+  }
+};
+
 template <typename T>
-using XPUABSFunctor = XPUActivationFunc<T, xpu::Activation_t::ABS>;
+struct XPULeakyReluGradFunctor : public BaseActivationFunctor<T> {
+  void operator()(const framework::ExecutionContext &ctx) const {
+    const auto *x = ctx.Input<Tensor>("X");
+    auto *dOut = ctx.Input<framework::Tensor>(framework::GradVarName("Out"));
+    auto *dX = ctx.Output<framework::Tensor>(framework::GradVarName("X"));
+    float alpha = ctx.Attr<float>("alpha");
+    const T *x_data = nullptr;
+    const T *y_grad = nullptr;
+    if (x != nullptr) x_data = x->data<T>();
+    if (dOut != nullptr) y_grad = dOut->data<T>();
+    T *x_grad = dX->mutable_data<T>(ctx.GetPlace());
+    auto xpu_context =
+        ctx.device_context<paddle::platform::XPUDeviceContext>().x_context();
+
+    // The signs of x and y are the same,
+    // y == nullptr here,
+    // so we give 2 x to the api
+    int r = xpu::leaky_relu_grad(
+        xpu_context, reinterpret_cast<const float *>(x_data),
+        reinterpret_cast<const float *>(x_data),
+        reinterpret_cast<const float *>(y_grad),
+        reinterpret_cast<float *>(x_grad), dX->numel(), alpha);
+    PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
+                      platform::errors::External(
+                          "XPU leaky_relu_grad return wrong value[%d %s].", r,
+                          XPUAPIErrorMsg[r]));
+  }
+};
+
 }  // namespace operators
 }  // namespace paddle
 
@@ -226,14 +347,16 @@ REGISTER_ACTIVATION_XPU_KERNEL(sigmoid, XPUSigmoidFunctor,
                                XPUSigmoidGradFunctor)
 REGISTER_ACTIVATION_XPU_KERNEL(gelu, XPUGeluFunctor, XPUGeluGradFunctor)
 REGISTER_ACTIVATION_XPU_KERNEL(sqrt, XPUSqrtFunctor, XPUSqrtGradFunctor)
-REGISTER_ACTIVATION_XPU_KERNEL(square, XPUSquareFunctor, XPUSuareGradFunctor)
+REGISTER_ACTIVATION_XPU_KERNEL(square, XPUSquareFunctor, XPUSquareGradFunctor)
 REGISTER_ACTIVATION_XPU_KERNEL(hard_swish, XPUHardSwishFunctor,
                                XPUHardSwishGradFunctor)
+REGISTER_ACTIVATION_XPU_KERNEL(leaky_relu, XPULeakyReluFunctor,
+                               XPULeakyReluGradFunctor)
 REGISTER_OP_XPU_KERNEL(log,
                        ops::XPUActivationKernel<ops::XPULogFunctor<float>>);
 REGISTER_OP_XPU_KERNEL(pow,
-                       ops::XPUActivationKernel<ops::XPUACTPowFunctor<float>>);
+                       ops::XPUActivationKernel<ops::XPUPowFunctor<float>>);
 REGISTER_OP_XPU_KERNEL(abs,
-                       ops::XPUActivationKernel<ops::XPUABSFunctor<float>>);
+                       ops::XPUActivationKernel<ops::XPUAbsFunctor<float>>);
 
 #endif  // PADDLE_WITH_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()

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

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import unittest
+import numpy as np
+import paddle
+import paddle.fluid.core as core
+import sys
+sys.path.append("..")
+from op_test_xpu import XPUOpTest
+import paddle.fluid as fluid
+from paddle.fluid import Program, program_guard
+
+paddle.enable_static()
+
+
+def nearest_neighbor_interp_np(X,
+                               out_h,
+                               out_w,
+                               out_size=None,
+                               actual_shape=None,
+                               align_corners=True,
+                               data_layout='NCHW'):
+    """nearest neighbor interpolation implement in shape [N, C, H, W]"""
+    if data_layout == "NHWC":
+        X = np.transpose(X, (0, 3, 1, 2))  # NHWC => NCHW
+    if out_size is not None:
+        out_h = out_size[0]
+        out_w = out_size[1]
+    if actual_shape is not None:
+        out_h = actual_shape[0]
+        out_w = actual_shape[1]
+    n, c, in_h, in_w = X.shape
+
+    ratio_h = ratio_w = 0.0
+    if (out_h > 1):
+        if (align_corners):
+            ratio_h = (in_h - 1.0) / (out_h - 1.0)
+        else:
+            ratio_h = 1.0 * in_h / out_h
+    if (out_w > 1):
+        if (align_corners):
+            ratio_w = (in_w - 1.0) / (out_w - 1.0)
+        else:
+            ratio_w = 1.0 * in_w / out_w
+
+    out = np.zeros((n, c, out_h, out_w))
+
+    if align_corners:
+        for i in range(out_h):
+            in_i = int(ratio_h * i + 0.5)
+            for j in range(out_w):
+                in_j = int(ratio_w * j + 0.5)
+                out[:, :, i, j] = X[:, :, in_i, in_j]
+    else:
+        for i in range(out_h):
+            in_i = int(ratio_h * i)
+            for j in range(out_w):
+                in_j = int(ratio_w * j)
+                out[:, :, i, j] = X[:, :, in_i, in_j]
+
+    if data_layout == "NHWC":
+        out = np.transpose(out, (0, 2, 3, 1))  # NCHW => NHWC
+
+    return out.astype(X.dtype)
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterpOp(XPUOpTest):
+    def setUp(self):
+        self.use_xpu = True
+        self.out_size = None
+        self.actual_shape = None
+        self.data_layout = 'NCHW'
+        self.init_test_case()
+        self.op_type = "nearest_interp"
+        input_np = np.random.random(self.input_shape).astype("float32")
+
+        if self.data_layout == "NCHW":
+            in_h = self.input_shape[2]
+            in_w = self.input_shape[3]
+        else:
+            in_h = self.input_shape[1]
+            in_w = self.input_shape[2]
+
+        if self.scale > 0:
+            out_h = int(in_h * self.scale)
+            out_w = int(in_w * self.scale)
+        else:
+            out_h = self.out_h
+            out_w = self.out_w
+
+        output_np = nearest_neighbor_interp_np(
+            input_np, out_h, out_w, self.out_size, self.actual_shape,
+            self.align_corners, self.data_layout)
+        self.inputs = {'X': input_np}
+        if self.out_size is not None:
+            self.inputs['OutSize'] = self.out_size
+        if self.actual_shape is not None:
+            self.inputs['OutSize'] = self.actual_shape
+        self.attrs = {
+            'out_h': self.out_h,
+            'out_w': self.out_w,
+            'scale': self.scale,
+            'interp_method': self.interp_method,
+            'align_corners': self.align_corners,
+            'data_layout': self.data_layout
+        }
+        self.outputs = {'Out': output_np}
+
+    def test_check_output(self):
+        place = paddle.XPUPlace(0)
+        self.check_output_with_place(place)
+
+    def test_check_grad(self):
+        place = paddle.XPUPlace(0)
+        self.check_grad_with_place(place, ['X'], 'Out', in_place=True)
+
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [2, 3, 4, 5]
+        self.out_h = 2
+        self.out_w = 2
+        self.scale = 0.
+        self.out_size = np.array([3, 3]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpCase1(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [4, 1, 7, 8]
+        self.out_h = 1
+        self.out_w = 1
+        self.scale = 0.
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpCase2(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 3, 9, 6]
+        self.out_h = 12
+        self.out_w = 12
+        self.scale = 0.
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpCase3(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [1, 1, 32, 64]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 0.
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpCase4(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [4, 1, 7, 8]
+        self.out_h = 1
+        self.out_w = 1
+        self.scale = 0.
+        self.out_size = np.array([2, 2]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpCase5(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 3, 9, 6]
+        self.out_h = 12
+        self.out_w = 12
+        self.scale = 0.
+        self.out_size = np.array([11, 11]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpCase6(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [1, 1, 32, 64]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 0.
+        self.out_size = np.array([65, 129]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpSame(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [2, 3, 32, 64]
+        self.out_h = 32
+        self.out_w = 64
+        self.scale = 0.
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpActualShape(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 2, 32, 16]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 0.
+        self.out_size = np.array([66, 40]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpDataLayout(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [2, 4, 4, 5]
+        self.out_h = 2
+        self.out_w = 2
+        self.scale = 0.
+        self.out_size = np.array([3, 8]).astype("int32")
+        self.align_corners = True
+        self.data_layout = "NCHW"
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterpWithoutCorners(TestNearestInterpOp):
+    def set_align_corners(self):
+        self.align_corners = False
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpScale1(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 2, 7, 5]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 2.
+        self.out_size = np.array([66, 40]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpScale2(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 2, 5, 7]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 1.5
+        self.out_size = np.array([66, 40]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestNeighborInterpScale3(TestNearestInterpOp):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 2, 7, 5]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 1.
+        self.out_size = np.array([66, 40]).astype("int32")
+        self.align_corners = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterpOp_attr_tensor(XPUOpTest):
+    def setUp(self):
+        self.out_size = None
+        self.actual_shape = None
+        self.init_test_case()
+        self.op_type = "nearest_interp"
+        self.shape_by_1Dtensor = False
+        self.scale_by_1Dtensor = False
+        self.attrs = {
+            'interp_method': self.interp_method,
+            'align_corners': self.align_corners,
+        }
+
+        input_np = np.random.random(self.input_shape).astype("float32")
+        self.inputs = {'X': input_np}
+
+        if self.scale_by_1Dtensor:
+            self.inputs['Scale'] = np.array([self.scale]).astype("float32")
+        elif self.scale > 0:
+            out_h = int(self.input_shape[2] * self.scale)
+            out_w = int(self.input_shape[3] * self.scale)
+            self.attrs['scale'] = self.scale
+        else:
+            out_h = self.out_h
+            out_w = self.out_w
+
+        if self.shape_by_1Dtensor:
+            self.inputs['OutSize'] = self.out_size
+        elif self.out_size is not None:
+            size_tensor = []
+            for index, ele in enumerate(self.out_size):
+                size_tensor.append(("x" + str(index), np.ones(
+                    (1)).astype('int32') * ele))
+            self.inputs['SizeTensor'] = size_tensor
+
+        self.attrs['out_h'] = self.out_h
+        self.attrs['out_w'] = self.out_w
+        output_np = nearest_neighbor_interp_np(input_np, out_h, out_w,
+                                               self.out_size, self.actual_shape,
+                                               self.align_corners)
+        self.outputs = {'Out': output_np}
+
+    def test_check_output(self):
+        place = paddle.XPUPlace(0)
+        self.check_output_with_place(place)
+
+    def test_check_grad(self):
+        place = paddle.XPUPlace(0)
+        self.check_grad_with_place(place, ['X'], 'Out', in_place=True)
+
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [2, 5, 4, 4]
+        self.out_h = 3
+        self.out_w = 3
+        self.scale = 0.
+        self.out_size = [3, 3]
+        self.align_corners = True
+
+
+# out_size is a tensor list
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterp_attr_tensor_Case1(TestNearestInterpOp_attr_tensor):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 3, 9, 6]
+        self.out_h = 12
+        self.out_w = 12
+        self.scale = 0.
+        self.out_size = [8, 12]
+        self.align_corners = True
+
+
+# out_size is a 1-D tensor
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterp_attr_tensor_Case2(TestNearestInterpOp_attr_tensor):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 2, 32, 16]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 0.
+        self.out_size = np.array([66, 40]).astype("int32")
+        self.align_corners = True
+        self.shape_by_1Dtensor = True
+
+
+# scale is a 1-D tensor
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterp_attr_tensor_Case3(TestNearestInterpOp_attr_tensor):
+    def init_test_case(self):
+        self.interp_method = 'nearest'
+        self.input_shape = [3, 2, 32, 16]
+        self.out_h = 64
+        self.out_w = 32
+        self.scale = 2.0
+        self.out_size = None
+        self.align_corners = True
+        self.scale_by_1Dtensor = True
+
+
+@unittest.skipIf(not paddle.is_compiled_with_xpu(),
+                 "core is not compiled with XPU")
+class TestNearestInterpException(unittest.TestCase):
+    def test_exception(self):
+        input = fluid.data(name="input", shape=[1, 3, 6, 6], dtype="float32")
+
+        def attr_data_format():
+            # for 4-D input, data_format can only be NCHW or NHWC
+            out = fluid.layers.resize_nearest(
+                input, out_shape=[4, 8], data_format='NDHWC')
+
+        def attr_scale_type():
+            out = fluid.layers.resize_nearest(input, scale='scale')
+
+        def attr_scale_value():
+            out = fluid.layers.resize_nearest(input, scale=-0.3)
+
+        self.assertRaises(ValueError, attr_data_format)
+        self.assertRaises(TypeError, attr_scale_type)
+        self.assertRaises(ValueError, attr_scale_value)
+
+
+if __name__ == "__main__":
+    unittest.main()