[XPU] add sampling_id op, add top_k op, update xdnn api

cmake/external/xpu.cmake

@@ -10,7 +10,7 @@ set(XPU_RT_LIB_NAME "libxpurt.so")
 if(NOT DEFINED XPU_BASE_URL)
   set(XPU_BASE_URL_WITHOUT_DATE
       "https://baidu-kunlun-product.cdn.bcebos.com/KL-SDK/klsdk-dev")
-  set(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220727")
+  set(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220728")
 else()
   set(XPU_BASE_URL "${XPU_BASE_URL}")
 endif()
@@ -19,7 +19,7 @@ endif()
 if(NOT DEFINED XPU_XDNN_BASE_URL)
   set(XPU_XDNN_BASE_URL_WITHOUT_DATE
       "https://klx-sdk-release-public.su.bcebos.com/xdnn/dev")
-  set(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL_WITHOUT_DATE}/20220727")
+  set(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL_WITHOUT_DATE}/20220728")
 else()
   set(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL}")
 endif()

paddle/fluid/framework/tensor_util.h

@@ -302,6 +302,11 @@ void TensorFromVector(const std::vector<T>& src,
         size,
         reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream());
   }
+#endif
+#ifdef PADDLE_WITH_XPU
+  else if (platform::is_xpu_place(dst_place)) {  // NOLINT
+    memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size);
+  }
 #endif
   else {  // NOLINT
     PADDLE_THROW(platform::errors::Unimplemented(
@@ -381,6 +386,11 @@ inline void TensorFromVector(const std::vector<bool>& src,
         reinterpret_cast<const platform::CustomDeviceContext&>(ctx).stream();
     memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size, stream);
   }
+#endif
+#ifdef PADDLE_WITH_XPU
+  else if (platform::is_xpu_place(dst_place)) {  // NOLINT
+    memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size);
+  }
 #endif
   else {  // NOLINT
     PADDLE_THROW(platform::errors::Unimplemented(

paddle/fluid/operators/detection/generate_proposals_v2_op_xpu.cc

@@ -219,20 +219,14 @@ static std::pair<Tensor, Tensor> ProposalForOneImage(
 
   // 4. nms
   int nms_keep_num = 0;
-  r = xpu::nms<T>(dev_ctx.x_context(),
-                  proposals_filter.data<T>(),
-                  nullptr,
-                  keep_index.data<int>(),
-                  1,
-                  1,
-                  keep_num,
-                  -1,
-                  nms_thresh,
-                  -1,
-                  0,
-                  &nms_keep_num,
-                  pixel_offset);
-  PADDLE_ENFORCE_XDNN_SUCCESS(r, "nms");
+  r = xpu::sorted_nms<T>(dev_ctx.x_context(),
+                         proposals_filter.data<T>(),
+                         keep_index.data<int>(),
+                         nms_keep_num,
+                         keep_num,
+                         nms_thresh,
+                         pixel_offset);
+  PADDLE_ENFORCE_XDNN_SUCCESS(r, "sorted_nms");
   if (post_nms_top_n > 0 && post_nms_top_n < nms_keep_num) {
     keep_index.Resize({post_nms_top_n});
   } else {

paddle/fluid/operators/one_hot_op_xpu.cc

@@ -17,6 +17,7 @@
 #include <vector>
 
 #include "paddle/fluid/operators/one_hot_op.h"
+#include "paddle/fluid/platform/device/device_wrapper.h"
 
 namespace paddle {
 namespace operators {
@@ -28,9 +29,13 @@ template <typename DeviceContext, typename T>
 class OneHotXPUKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
-    auto* in = context.Input<LoDTensor>("X");
+    const auto* in = context.Input<LoDTensor>("X");
     auto* out = context.Output<LoDTensor>("Out");
+
+    // get depth from attr
     int depth = context.Attr<int>("depth");
+
+    // get depth from input tensor
     if (context.HasInput("depth_tensor")) {
       auto* depth_tensor = context.Input<Tensor>("depth_tensor");
       auto* depth_data = depth_tensor->data<int32_t>();
@@ -50,18 +55,14 @@ class OneHotXPUKernel : public framework::OpKernel<T> {
 
     auto& dev_ctx = context.template device_context<DeviceContext>();
     int len = in->numel();
+    // int one_hot(Context* ctx, const T* x, float* y, int len, int depth, float
+    // on_value = 1.0f, float off_value = 0.0f);
     int ret = xpu::one_hot<T>(dev_ctx.x_context(),
                               in->data<T>(),
                               out->mutable_data<float>(context.GetPlace()),
                               len,
                               depth);
-
-    PADDLE_ENFORCE_EQ(ret,
-                      XPU_SUCCESS,
-                      platform::errors::External(
-                          "XPU one_hot kernel return wrong value[%d %s]",
-                          ret,
-                          XPUAPIErrorMsg[ret]));
+    PADDLE_ENFORCE_XDNN_SUCCESS(ret, "one_hot");
   }
 };
 

paddle/fluid/operators/sampling_id_op_xpu.cc

@@ -0,0 +1,20 @@
+/* Copyright (c) 2022 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. */
+
+#include "paddle/fluid/operators/sampling_id_op.h"
+
+namespace ops = paddle::operators;
+REGISTER_OP_XPU_KERNEL(sampling_id,
+                       paddle::operators::SamplingIdKernel<float>,
+                       paddle::operators::SamplingIdKernel<double>);

paddle/fluid/platform/device/xpu/xpu2_op_list.h

@@ -322,6 +322,9 @@ XPUOpMap& get_kl2_ops() {
        XPUKernelSet({pOpKernelType(vartype::INT64, XPUPlace()),
                      pOpKernelType(vartype::INT32, XPUPlace()),
                      pOpKernelType(vartype::FP32, XPUPlace())})},
+      {"one_hot",
+       XPUKernelSet({pOpKernelType(vartype::INT32, XPUPlace()),
+                     pOpKernelType(vartype::INT64, XPUPlace())})},
       {"one_hot_v2",
        XPUKernelSet({pOpKernelType(vartype::INT32, XPUPlace()),
                      pOpKernelType(vartype::INT64, XPUPlace())})},
@@ -393,6 +396,9 @@ XPUOpMap& get_kl2_ops() {
       {"scatter",
        XPUKernelSet({pOpKernelType(vartype::INT64, XPUPlace()),
                      pOpKernelType(vartype::FP32, XPUPlace())})},
+      {"sampling_id",
+       XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace()),
+                     pOpKernelType(vartype::FP64, XPUPlace())})},
       {"sgd",
        XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace()),
                      pOpKernelType(vartype::FP16, XPUPlace())})},