PaddlePaddle · chenwhql · Mar 11, 2022 · Mar 8, 2022 · Mar 8, 2022 · Mar 8, 2022
diff --git a/paddle/fluid/distributed/fleet_executor/test/compute_interceptor_run_op_test.cc b/paddle/fluid/distributed/fleet_executor/test/compute_interceptor_run_op_test.cc
@@ -24,10 +24,14 @@ limitations under the License. */
 #include "paddle/fluid/distributed/fleet_executor/task_node.h"
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/program_desc.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP_ITSELF(elementwise_add);
 USE_OP_ITSELF(fill_constant);
 
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace distributed {
 

diff --git a/paddle/fluid/eager/tests/data_structure_tests/grad_tensor_holder_test.cc b/paddle/fluid/eager/tests/data_structure_tests/grad_tensor_holder_test.cc
@@ -24,6 +24,8 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full_like, CPU, ALL_LAYOUT);
+
 // TODO(jiabin): remove nolint here!!!
 using namespace egr;  // NOLINT
 

diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
@@ -33,6 +33,14 @@
 #include "gperftools/profiler.h"
 #endif
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_grad, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
+
 using namespace egr;            // NOLINT
 using namespace egr_utils_api;  // NOLINT
 

diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
@@ -32,11 +32,19 @@
 #include "gperftools/profiler.h"
 #endif
 
+#include "paddle/phi/core/kernel_registry.h"
+
 using namespace egr;            // NOLINT
 using namespace egr_utils_api;  // NOLINT
 
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 
+PD_DECLARE_KERNEL(full, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_grad, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, GPU, ALL_LAYOUT);
+
 TEST(Benchmark, EagerScaleCUDA) {
   eager_test::InitEnv(paddle::platform::CUDAPlace());
 

diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
@@ -34,6 +34,14 @@
 #include "gperftools/profiler.h"
 #endif
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_grad, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace imperative {
 

diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
@@ -34,8 +34,16 @@
 #include "gperftools/profiler.h"
 #endif
 
+#include "paddle/phi/core/kernel_registry.h"
+
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 
+PD_DECLARE_KERNEL(full, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_grad, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, GPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace imperative {
 

diff --git a/paddle/fluid/eager/tests/task_tests/backward_test.cc b/paddle/fluid/eager/tests/task_tests/backward_test.cc
@@ -30,6 +30,10 @@
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/tensor_meta.h"
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 TEST(Backward, SingleNodeEmptyGrad) {

diff --git a/paddle/fluid/eager/tests/task_tests/cross_batch_accumulation_test.cc b/paddle/fluid/eager/tests/task_tests/cross_batch_accumulation_test.cc
@@ -31,6 +31,10 @@
 
 #include "paddle/fluid/eager/tests/test_utils.h"
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 TEST(CrossBatchAccumulation, SingleScaleNode) {

diff --git a/paddle/fluid/eager/tests/task_tests/forward_autograd_test.cc b/paddle/fluid/eager/tests/task_tests/forward_autograd_test.cc
@@ -27,6 +27,10 @@
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/tensor_meta.h"
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 TEST(Forward, SingleNode) {

diff --git a/paddle/fluid/eager/tests/task_tests/fwd_bwd_joint_test.cc b/paddle/fluid/eager/tests/task_tests/fwd_bwd_joint_test.cc
@@ -30,6 +30,13 @@
 #include "paddle/fluid/eager/hooks.h"
 #include "paddle/fluid/eager/tests/test_utils.h"
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(full, GPU, ALL_LAYOUT);
+#endif
+
 namespace egr {
 
 paddle::experimental::Tensor hook_function(

diff --git a/paddle/fluid/eager/tests/task_tests/generated_test.cc b/paddle/fluid/eager/tests/task_tests/generated_test.cc
@@ -30,6 +30,12 @@
 #include "paddle/fluid/eager/api/generated/fluid_generated/dygraph_forward_api.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_grad, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 TEST(Generated, Sigmoid) {

diff --git a/paddle/fluid/eager/tests/task_tests/hook_test.cc b/paddle/fluid/eager/tests/task_tests/hook_test.cc
@@ -31,6 +31,10 @@
 #include "paddle/fluid/eager/hooks.h"
 #include "paddle/fluid/eager/tests/test_utils.h"
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 paddle::experimental::Tensor hook_function(

diff --git a/paddle/fluid/eager/tests/task_tests/hook_test_intermidiate.cc b/paddle/fluid/eager/tests/task_tests/hook_test_intermidiate.cc
@@ -27,6 +27,12 @@
 #include "paddle/fluid/eager/hooks.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_grad, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 paddle::experimental::Tensor hook_function(

diff --git a/paddle/fluid/eager/tests/task_tests/tensor_utils_test.cc b/paddle/fluid/eager/tests/task_tests/tensor_utils_test.cc
@@ -23,6 +23,10 @@
 #include "paddle/fluid/eager/tests/test_utils.h"
 #include "paddle/phi/api/lib/utils/allocator.h"
 
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace egr {
 
 TEST(TensorUtils, Test) {

diff --git a/paddle/fluid/framework/ir/memory_optimize_pass/test_reference_count_pass_last_lived_ops.cc b/paddle/fluid/framework/ir/memory_optimize_pass/test_reference_count_pass_last_lived_ops.cc
@@ -20,12 +20,15 @@
 #include "paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass_helper.h"
 #include "paddle/fluid/framework/parallel_executor.h"
 #include "paddle/fluid/framework/program_desc.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP_ITSELF(scale);
 USE_OP(elementwise_mul);
 USE_OP_ITSELF(elementwise_add);
 USE_OP_ITSELF(elementwise_add_grad);
 
+PD_DECLARE_KERNEL(scale, CPU, ALL_LAYOUT);
+
 DECLARE_double(eager_delete_tensor_gb);
 
 namespace paddle {

diff --git a/paddle/fluid/imperative/tests/test_hooks.cc b/paddle/fluid/imperative/tests/test_hooks.cc
@@ -24,6 +24,10 @@
 #include "paddle/fluid/imperative/hooks.h"
 #include "paddle/fluid/imperative/tracer.h"
 #include "paddle/fluid/memory/memcpy.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
 
 namespace platform = paddle::platform;
 namespace framework = paddle::framework;

diff --git a/paddle/fluid/imperative/tests/test_prepare_op.cc b/paddle/fluid/imperative/tests/test_prepare_op.cc
@@ -24,6 +24,13 @@
 #include "paddle/fluid/framework/op_info.h"
 #include "paddle/fluid/imperative/prepared_operator.h"
 #include "paddle/fluid/imperative/type_defs.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(split, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(relu, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(relu, GPU, ALL_LAYOUT);
+#endif
 
 namespace imperative = paddle::imperative;
 namespace platform = paddle::platform;

diff --git a/paddle/fluid/imperative/tests/test_tracer.cc b/paddle/fluid/imperative/tests/test_tracer.cc
@@ -28,6 +28,13 @@
 #include "paddle/fluid/imperative/tracer.h"
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/fluid/platform/device_context.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(add_grad, GPU, ALL_LAYOUT);
+#endif
 
 namespace imperative = paddle::imperative;
 namespace platform = paddle::platform;

diff --git a/paddle/fluid/operators/cinn/cinn_instruction_run_op_test.cc b/paddle/fluid/operators/cinn/cinn_instruction_run_op_test.cc
@@ -22,11 +22,17 @@ limitations under the License. */
 #include "paddle/fluid/platform/cpu_helper.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/init.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP(cinn_launch);
 USE_OP(cinn_instruction_run);
 USE_OP_ITSELF(elementwise_add);
 
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+#ifdef PADDLE_WITH_CUDA
+PD_DECLARE_KERNEL(add, GPU, ALL_LAYOUT);
+#endif
+
 namespace paddle::operators {
 
 using framework::paddle2cinn::CinnCompiler;

diff --git a/paddle/fluid/operators/cinn/cinn_launch_op_test.cc b/paddle/fluid/operators/cinn/cinn_launch_op_test.cc
@@ -26,12 +26,18 @@ limitations under the License. */
 #include "paddle/fluid/platform/cpu_helper.h"
 #include "paddle/fluid/platform/init.h"
 #include "paddle/phi/core/ddim.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP(cinn_launch);
 USE_OP(cinn_instruction_run);
 USE_OP_ITSELF(elementwise_add);
 DECLARE_double(eager_delete_tensor_gb);
 
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+#ifdef PADDLE_WITH_CUDA
+PD_DECLARE_KERNEL(add, GPU, ALL_LAYOUT);
+#endif
+
 namespace paddle::operators {
 
 using framework::paddle2cinn::CinnCompiler;

diff --git a/paddle/fluid/operators/elementwise/test_elementwise_add_grad_grad.cc b/paddle/fluid/operators/elementwise/test_elementwise_add_grad_grad.cc
@@ -17,8 +17,13 @@
 #include "paddle/fluid/operators/elementwise/test_elementwise_op_grad_grad.h"
 #include "paddle/fluid/platform/place.h"
 #include "paddle/phi/core/ddim.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP_ITSELF(elementwise_add);
+PD_DECLARE_KERNEL(add_double_grad, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(add_double_grad, GPU, ALL_LAYOUT);
+#endif
 
 namespace paddle {
 namespace operators {

diff --git a/paddle/fluid/operators/elementwise/test_elementwise_add_op_inplace.cc b/paddle/fluid/operators/elementwise/test_elementwise_add_op_inplace.cc
@@ -21,9 +21,12 @@
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/place.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP_ITSELF(elementwise_add);
 
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace operators {
 

diff --git a/paddle/fluid/operators/elementwise/test_elementwise_div_grad_grad.cc b/paddle/fluid/operators/elementwise/test_elementwise_div_grad_grad.cc
@@ -27,9 +27,15 @@
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/place.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP_ITSELF(elementwise_div);
 
+PD_DECLARE_KERNEL(divide_double_grad, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(divide_double_grad, GPU, ALL_LAYOUT);
+#endif
+
 namespace paddle {
 namespace operators {
 

diff --git a/paddle/fluid/operators/feed_forward_test.cu b/paddle/fluid/operators/feed_forward_test.cu
@@ -21,6 +21,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/fused/attn_feed_forward.h"
 #include "paddle/fluid/platform/float16.h"
+#include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 namespace framework = paddle::framework;
@@ -29,6 +30,11 @@ namespace platform = paddle::platform;
 USE_OP(matmul);
 USE_OP_ITSELF(elementwise_add);
 
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(add_grad, GPU, ALL_LAYOUT);
+#endif
+
 // get paddle matmul op results as baseline
 template <typename T>
 void GetLinearOp(const std::vector<T> &x, const std::vector<T> &y,

diff --git a/paddle/fluid/operators/fused/fused_dropout_act_bias_test.cu b/paddle/fluid/operators/fused/fused_dropout_act_bias_test.cu
@@ -20,8 +20,14 @@ limitations under the License. */
 #include "paddle/fluid/operators/amp/fp16_type_traits.h"
 #include "paddle/fluid/operators/fused/fused_dropout_act_bias.h"
 #include "paddle/fluid/operators/fused/fused_dropout_test.h"
+#include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/functors.h"
 
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(dropout, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(dropout_grad, GPU, ALL_LAYOUT);
+#endif
+
 namespace framework = paddle::framework;
 namespace platform = paddle::platform;
 namespace details = paddle::operators::details;

diff --git a/paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias_test.cu b/paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias_test.cu
@@ -19,6 +19,12 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/fused/fused_dropout_test.h"
 #include "paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(dropout, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(dropout_grad, GPU, ALL_LAYOUT);
+#endif
 
 /**
  * @brief The unit test of fused_layernorm_residual_dropout_bias

diff --git a/paddle/fluid/operators/fused/fused_residual_dropout_bias_test.cu b/paddle/fluid/operators/fused/fused_residual_dropout_bias_test.cu
@@ -19,6 +19,12 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/fused/fused_dropout_test.h"
 #include "paddle/fluid/operators/fused/fused_residual_dropout_bias.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(dropout, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(dropout_grad, GPU, ALL_LAYOUT);
+#endif
 
 namespace framework = paddle::framework;
 namespace platform = paddle::platform;

diff --git a/paddle/fluid/operators/mkldnn/test_mkldnn_op_nhwc.cc b/paddle/fluid/operators/mkldnn/test_mkldnn_op_nhwc.cc
@@ -24,6 +24,7 @@
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/place.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP(pool2d);
 USE_OP_DEVICE_KERNEL(pool2d, MKLDNN);
@@ -32,6 +33,8 @@ USE_OP_DEVICE_KERNEL(relu, MKLDNN);
 USE_OP_ITSELF(transpose);
 USE_OP_DEVICE_KERNEL(transpose, MKLDNN);
 
+PD_DECLARE_KERNEL(relu, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace operators {