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
index 18920d06f38543cc3f7aeb045e7c3058143e006e..ba039385a74ba45aa1f33ba38138d8e5213f2e00 100644
--- 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
index 8c6eeca9d3d5d80fd5bfe943ef87ba8640ada4f2..384fdcd6f97c4b318341db68cdd88b644d42d22a 100644
--- 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
index 6c4bf9a4f17e6f88503f0a1d6ec2f3029000b6f0..af365322e606ebfaecb7233751cacc6aa1aac423 100644
--- 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
index 9f59f4fc03045aa8a122820e2328a76f33ad9877..5b75f1242e69bc5b37dd97467b7c55bfc6bc3871 100644
--- 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
index 3292de9363696dae30d853980eca6fb1ba1055cc..a9d297c1c64f7b64373237a0500802a5c883aedd 100644
--- 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
index df77fc1360b4994310ddce242459b435d771ee20..bd9eaa09ca9a406da943c8a0b0f37b674d5ea3c2 100644
--- 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
index a4bc56bd606f3fbb0f9152d58acb5c8edeecf905..0c894ed267fcdd08d44d4df08bfaf0554874aebf 100644
--- 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
index 524872b2e55638d25697388aa50724f49f6e3818..36594f1aac8cdb131bb77f1396dca19a0c2e8cc0 100644
--- 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
index 49bbfc77741a5b82ac9a564e25b484e5dabf77a7..dc44d95daac1d9109bbf2a1d04a8a47b081cead9 100644
--- 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
index 5a7bafb2fe37051c0ad054c130d77dd6e05319d2..f7fa642ea8dd17d20816e74c9bfb4cd92b184b4a 100644
--- 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
index 4b7077b13bdd6c48a0a3846656bd3a6337eb9f80..2a5ad53204a6201149bec0b3dac0fa3baf441f2e 100644
--- 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
index 9cda961741f55e9b4b7fc8dac61fe4a7c96567cf..d546df4ed087a99a28096a5336fab3826991534a 100644
--- 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
index 15b2a62dca751859882e82d46acaa46f27c2c518..56813c498d2410caa452da7a334c393b230c65bf 100644
--- 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
index ea821d195099f3d632e0d1b2d4937bac812563c8..24e5da060111f083ef9b65574e75295fa07f8f43 100644
--- 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
index d33dc7f49feb0f4c9e585d13186d65b6c2d618c0..636a594a657cb0744aac161d928ff9078b1f92bc 100644
--- 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
index 3ac2028790608529e0745dde2ce41ed57748f46d..02a1689c23a3fe5e1543a2e52d7661d5997bc062 100644
--- 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
index 17cbe06748234ae518ab2f55fc11c936e73c470e..4cda3f32fdf3fdd2d14b201fa902c1f50f3ff98d 100644
--- 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
index 0696de908a917cfd7b0ebd1334bb2ad4e12559ee..2e38bd77cf63cc85b75a50e62250a6e746f525bc 100644
--- 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
index 2afee35112e6f74df74a095c492f60cce0f9786c..0edbee534c0b5d680717250e7702f272eacd0272 100644
--- 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
index 460d417e61fd4c77dcc9d581a3b997e6ae7c8e4c..585f1caabed051134fd5ce7624c17b741b487ef0 100644
--- 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
index 5222103256d614a2d6b1fa10662367ecb20d3cb2..ea009a38056f078689bd6dc4c9a41d2b34e8c1fa 100644
--- 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
index 9d4d11609ac2047aa8934cb2868f79359a816e12..ce5c6b701d95894db8e3a84215f537352914706a 100644
--- 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
index 7890d634e9941718f3420bd50a7ded453379fc69..3cecc52a3c481cf9cb4a1e2eba6ded704a8fa8ee 100644
--- 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
index 0eb84f18f25f03b1fd0310c5815ee342ff835a6f..27a235765227f15dd412dcd6ad55f2a24471c6da 100644
--- 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
index 2381b5b7fdfb85cbaa3fd66a10c5b630bb515f15..717c1732b7b3acf8528887aae43471c0dc0716e3 100644
--- 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
index cc14d0680d381ff2bbe73ee712e218c9c4d79185..032440d7f0478dc087e3ba38274f2a31a9a66a23 100644
--- 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
index 1a12e6b565f02035b3fb9673636c2344823f288e..5dff5e2225f4f3bf3a20daa02b2b4194bd8cb99e 100644
--- 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
index 916f02179b364bb26cf3312300b0ef0a07c1433b..9d0062e31388413fd4a441687631faebe8846c6e 100644
--- 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 {
 
diff --git a/paddle/fluid/operators/op_debug_string_test.cc b/paddle/fluid/operators/op_debug_string_test.cc
index b96fcaa486cce8099cf1d03c7d948ea74c1923ad..372a71706ab5ec72b6da4cbac1b63333f42cb265 100644
--- a/paddle/fluid/operators/op_debug_string_test.cc
+++ b/paddle/fluid/operators/op_debug_string_test.cc
@@ -17,8 +17,10 @@
 #include "glog/logging.h"
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/op_registry.h"
+#include "paddle/phi/core/kernel_registry.h"
 
 USE_OP_ITSELF(elementwise_add_grad);
+PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
 
 namespace paddle {
 namespace operators {
diff --git a/paddle/phi/tests/api/CMakeLists.txt b/paddle/phi/tests/api/CMakeLists.txt
index be12960d1d675b46987996713a0631399d1f0652..d998ab9435c027afcae0446fe3d31b3f000f1d15 100644
--- a/paddle/phi/tests/api/CMakeLists.txt
+++ b/paddle/phi/tests/api/CMakeLists.txt
@@ -1,28 +1,29 @@
 if(WITH_ROCM)
-  hip_test(test_phi_tensor SRCS test_pten_tensor.cc DEPS phi_tensor phi_function_api glog)
+  hip_test(test_phi_tensor SRCS test_pten_tensor.cc DEPS phi_tensor glog)
 else()
-  cc_test(test_phi_tensor SRCS test_pten_tensor.cc DEPS phi_tensor phi_function_api glog)
+  cc_test(test_phi_tensor SRCS test_pten_tensor.cc DEPS phi_tensor glog)
 endif()
 
 cc_test(test_phi_exception SRCS test_pten_exception.cc DEPS gtest)
 
-cc_test(test_mean_api SRCS test_mean_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_dot_api SRCS test_dot_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_matmul_api SRCS test_matmul_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_empty_api SRCS test_empty_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_fill_api SRCS test_fill_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_flatten_api SRCS test_flatten_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_elementwise_api SRCS test_elementwise_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_cast_api SRCS test_cast_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_reshape_api SRCS test_reshape_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_to_api SRCS test_to_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_slice_api SRCS test_slice_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_sum_api SRCS test_sum_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_scale_api SRCS test_scale_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_scale_benchmark SRCS test_scale_benchmark.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_conj_api SRCS test_conj_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_concat_api SRCS test_concat_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_split_api SRCS test_split_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_data_transform SRCS test_data_transform.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_sparse_utils_api SRCS test_sparse_utils_api.cc DEPS phi_tensor phi_api phi_api_utils)
-cc_test(test_sparse_conv_api SRCS test_sparse_conv_api.cc DEPS phi_tensor phi_api phi_api_utils)
+set(COMMON_API_TEST_DEPS phi_tensor phi_api phi_api_utils)
+cc_test(test_mean_api SRCS test_mean_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_dot_api SRCS test_dot_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_matmul_api SRCS test_matmul_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_empty_api SRCS test_empty_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_fill_api SRCS test_fill_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_flatten_api SRCS test_flatten_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_elementwise_api SRCS test_elementwise_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_cast_api SRCS test_cast_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_reshape_api SRCS test_reshape_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_to_api SRCS test_to_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_slice_api SRCS test_slice_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_sum_api SRCS test_sum_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_scale_api SRCS test_scale_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_scale_benchmark SRCS test_scale_benchmark.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_conj_api SRCS test_conj_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_concat_api SRCS test_concat_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_split_api SRCS test_split_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_data_transform SRCS test_data_transform.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_sparse_utils_api SRCS test_sparse_utils_api.cc DEPS ${COMMON_API_TEST_DEPS})
+cc_test(test_sparse_conv_api SRCS test_sparse_conv_api.cc DEPS ${COMMON_API_TEST_DEPS})
diff --git a/paddle/phi/tests/api/test_cast_api.cc b/paddle/phi/tests/api/test_cast_api.cc
index 276a70066ba73f3831c2a4bab0b47566848599f5..5448fb9d424709c34dc2008ef1ad7042c02d6630 100644
--- a/paddle/phi/tests/api/test_cast_api.cc
+++ b/paddle/phi/tests/api/test_cast_api.cc
@@ -21,6 +21,9 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(cast, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_concat_api.cc b/paddle/phi/tests/api/test_concat_api.cc
index d5a36f56bfa1b34a23e8bc4738fcec04225858a3..824b72b97ac129cc0f011dc2bc7279b4c1b362a9 100644
--- a/paddle/phi/tests/api/test_concat_api.cc
+++ b/paddle/phi/tests/api/test_concat_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(concat, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_conj_api.cc b/paddle/phi/tests/api/test_conj_api.cc
index 9c438de9297fb675fa391e3f32cee4761631e5b6..62a588dff12804391df971681b959d84a9c3ca7b 100644
--- a/paddle/phi/tests/api/test_conj_api.cc
+++ b/paddle/phi/tests/api/test_conj_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(conj, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_data_transform.cc b/paddle/phi/tests/api/test_data_transform.cc
index a3c497bd427ae040b33dce241a70ecaafee5fbcc..dd008ff36d50a8e30166d478b916ac180d564c7c 100644
--- a/paddle/phi/tests/api/test_data_transform.cc
+++ b/paddle/phi/tests/api/test_data_transform.cc
@@ -19,6 +19,16 @@ limitations under the License. */
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/core/compat/convert_utils.h"
 #include "paddle/phi/core/dense_tensor.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, CPU, ALL_LAYOUT);
+
+#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(copy, GPU, ALL_LAYOUT);
+#endif
 
 namespace paddle {
 namespace tests {
diff --git a/paddle/phi/tests/api/test_dot_api.cc b/paddle/phi/tests/api/test_dot_api.cc
index e48004653d638121b2b7e6e7047d3dfb51cd76c1..3fcd4e8a01d126f2d4f0fbaaacb7580b2a6b3163 100644
--- a/paddle/phi/tests/api/test_dot_api.cc
+++ b/paddle/phi/tests/api/test_dot_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(dot, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_elementwise_api.cc b/paddle/phi/tests/api/test_elementwise_api.cc
index cebf4e003aafa928341481952606f18642881a1c..d4013a788c76cb4f049574ee893320088fe7ac2c 100644
--- a/paddle/phi/tests/api/test_elementwise_api.cc
+++ b/paddle/phi/tests/api/test_elementwise_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_empty_api.cc b/paddle/phi/tests/api/test_empty_api.cc
index dc5618f0aae8ad2df85ef6c466457a7c6b303ad5..48adbe1bd2682c3096cb76e24ddc8e2f719a532e 100644
--- a/paddle/phi/tests/api/test_empty_api.cc
+++ b/paddle/phi/tests/api/test_empty_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(empty, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_fill_api.cc b/paddle/phi/tests/api/test_fill_api.cc
index 9b434aef811956bc6697193947a4756689c5dba3..bf57574d39093cc81a47f8385097627eb98648af 100644
--- a/paddle/phi/tests/api/test_fill_api.cc
+++ b/paddle/phi/tests/api/test_fill_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_flatten_api.cc b/paddle/phi/tests/api/test_flatten_api.cc
index e1360e8e27bff08397031e15377a1cf4a04c4ec7..f1c8935e266409dfd4aa4f86d59a87c2767b4eb9 100644
--- a/paddle/phi/tests/api/test_flatten_api.cc
+++ b/paddle/phi/tests/api/test_flatten_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(flatten, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_matmul_api.cc b/paddle/phi/tests/api/test_matmul_api.cc
index 2a3dd9c7dff62071fcd7dcf18cddcc5946ff7480..e2c324a6775c8f697c5871dcfe10a830244a103c 100644
--- a/paddle/phi/tests/api/test_matmul_api.cc
+++ b/paddle/phi/tests/api/test_matmul_api.cc
@@ -26,6 +26,15 @@ limitations under the License. */
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/platform/device_context.h"
+
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_double_grad, CPU, ALL_LAYOUT);
+
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(matmul, GPU, ALL_LAYOUT);
+#endif
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_mean_api.cc b/paddle/phi/tests/api/test_mean_api.cc
index 53be1b1e9dc9c29bddad810d740d0f7f05d28aa0..af47f2cd7714a2431b62acdedc61fea5a139bc49 100644
--- a/paddle/phi/tests/api/test_mean_api.cc
+++ b/paddle/phi/tests/api/test_mean_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(mean, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_pten_tensor.cc b/paddle/phi/tests/api/test_pten_tensor.cc
index dc2883c1794e2c986ed5446981b749f5f4dd0bc2..74ed648f3ee6ec4e2b763f4a57cc58ce79fb25b9 100644
--- a/paddle/phi/tests/api/test_pten_tensor.cc
+++ b/paddle/phi/tests/api/test_pten_tensor.cc
@@ -16,6 +16,13 @@
 #include "gtest/gtest.h"
 #include "paddle/phi/api/include/tensor.h"
 #include "paddle/phi/api/lib/ext_compat_utils.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
+
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(copy, GPU, ALL_LAYOUT);
+#endif
 
 namespace paddle {
 namespace tests {
diff --git a/paddle/phi/tests/api/test_reshape_api.cc b/paddle/phi/tests/api/test_reshape_api.cc
index 60281a9f4992344e19b8a05a3002d964695de94c..4a857e2d1dcda906ec2be5f1d9a20b51dbb53b22 100644
--- a/paddle/phi/tests/api/test_reshape_api.cc
+++ b/paddle/phi/tests/api/test_reshape_api.cc
@@ -21,6 +21,9 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(reshape, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_scale_api.cc b/paddle/phi/tests/api/test_scale_api.cc
index 52e8ae630e0e535e68e5fc9b76c1e9aca01ddd2c..a40ecc8485e4a96e96f40d513a153df4cc2eb0c3 100644
--- a/paddle/phi/tests/api/test_scale_api.cc
+++ b/paddle/phi/tests/api/test_scale_api.cc
@@ -19,8 +19,13 @@ limitations under the License. */
 
 #include "paddle/phi/api/lib/utils/allocator.h"
 #include "paddle/phi/core/dense_tensor.h"
+#include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/selected_rows.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(scale, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(scale_sr, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_scale_benchmark.cc b/paddle/phi/tests/api/test_scale_benchmark.cc
index 9c0b0fc576ebc7363d9c726a759f077a212c7eb8..05a5563344966e4d0664a83051dc84b3f2da499a 100644
--- a/paddle/phi/tests/api/test_scale_benchmark.cc
+++ b/paddle/phi/tests/api/test_scale_benchmark.cc
@@ -19,9 +19,12 @@ limitations under the License. */
 
 #include "paddle/phi/api/lib/utils/allocator.h"
 #include "paddle/phi/core/dense_tensor.h"
+#include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/tests/api/scale_api.h"
 #include "paddle/phi/tests/core/timer.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_slice_api.cc b/paddle/phi/tests/api/test_slice_api.cc
index c3f5fdcb36d62c926c6afc65bcc1ed9327983c69..ee2ade0229f1f3c9b59b87c2ffb37cf42ec1b531 100644
--- a/paddle/phi/tests/api/test_slice_api.cc
+++ b/paddle/phi/tests/api/test_slice_api.cc
@@ -19,6 +19,8 @@ limitations under the License. */
 #include "paddle/phi/api/include/tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_sparse_conv_api.cc b/paddle/phi/tests/api/test_sparse_conv_api.cc
index 16d7cb66f4cc5f14abf31bb0a16d58c266bc15fb..76cb01d8a8b98b070d89ba4a3887275f00e228f3 100644
--- a/paddle/phi/tests/api/test_sparse_conv_api.cc
+++ b/paddle/phi/tests/api/test_sparse_conv_api.cc
@@ -24,6 +24,8 @@ limitations under the License. */
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/sparse_coo_tensor.h"
 
+PD_DECLARE_KERNEL(sparse_conv3d, CPU, ALL_LAYOUT);
+
 template <typename T>
 void TestConv3dBase(const std::vector<int>& indices,
                     const std::vector<T>& features,
diff --git a/paddle/phi/tests/api/test_sparse_utils_api.cc b/paddle/phi/tests/api/test_sparse_utils_api.cc
index 819122a9b36650f8aa793b7d47ac2c0b6feb8f8e..8595782be35ab677ef40eb31b8a09237e90f359a 100644
--- a/paddle/phi/tests/api/test_sparse_utils_api.cc
+++ b/paddle/phi/tests/api/test_sparse_utils_api.cc
@@ -24,6 +24,8 @@ limitations under the License. */
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/sparse_coo_tensor.h"
 
+PD_DECLARE_KERNEL(dense_to_sparse_coo, CPU, ALL_LAYOUT);
+
 TEST(API, to_sparse_coo) {
   const auto alloc = std::make_shared<paddle::experimental::DefaultAllocator>(
       paddle::platform::CPUPlace());
diff --git a/paddle/phi/tests/api/test_split_api.cc b/paddle/phi/tests/api/test_split_api.cc
index 0b836a010586d775ced1e7c196b1b5139ac42fc1..1b84e7793cf6a864aaa3d9aa5ae7be8144e68c07 100644
--- a/paddle/phi/tests/api/test_split_api.cc
+++ b/paddle/phi/tests/api/test_split_api.cc
@@ -21,6 +21,8 @@
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(split, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_sum_api.cc b/paddle/phi/tests/api/test_sum_api.cc
index 80620b8e61c57668c8b090af798fa04c0da4779b..9781d70d2b9136071d4dde9a947af14f8d8b8e87 100644
--- a/paddle/phi/tests/api/test_sum_api.cc
+++ b/paddle/phi/tests/api/test_sum_api.cc
@@ -21,6 +21,8 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(sum, CPU, ALL_LAYOUT);
+
 namespace paddle {
 namespace tests {
 
diff --git a/paddle/phi/tests/api/test_to_api.cc b/paddle/phi/tests/api/test_to_api.cc
index d337a0b601a00d6ae0423b7184d4d2c5cc6ef2b8..66c478e4c000150b8bfd50d6cb5b79e90d7a5d80 100644
--- a/paddle/phi/tests/api/test_to_api.cc
+++ b/paddle/phi/tests/api/test_to_api.cc
@@ -21,6 +21,11 @@ limitations under the License. */
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
 
+PD_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PD_DECLARE_KERNEL(copy, GPU, ALL_LAYOUT);
+#endif
+
 namespace paddle {
 namespace tests {
 
diff --git a/python/paddle/utils/code_gen/api_gen.py b/python/paddle/utils/code_gen/api_gen.py
index 058cc08465ff048655de9d013adf206ea7ced055..a404fc01784154900a7c6ac1df501424dcdb307e 100644
--- a/python/paddle/utils/code_gen/api_gen.py
+++ b/python/paddle/utils/code_gen/api_gen.py
@@ -147,7 +147,6 @@ def source_include(header_file_path):
 #include "paddle/phi/infermeta/multiary.h"
 #include "paddle/phi/infermeta/nullary.h"
 #include "paddle/phi/infermeta/unary.h"
-#include "paddle/phi/kernels/declarations.h"
 """
 
 
diff --git a/python/paddle/utils/code_gen/sparse_api_gen.py b/python/paddle/utils/code_gen/sparse_api_gen.py
index 3838ac01c747c30797b0c7e8aabf5df677af097f..dd22e16dc64f0a12c3575b6f2a0d2c21cd97955b 100644
--- a/python/paddle/utils/code_gen/sparse_api_gen.py
+++ b/python/paddle/utils/code_gen/sparse_api_gen.py
@@ -188,7 +188,6 @@ def source_include(header_file_path):
 #include "paddle/phi/api/lib/kernel_dispatch.h"
 #include "paddle/phi/api/lib/sparse_api_custom_impl.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/declarations.h"
 """
 
 
diff --git a/python/paddle/utils/code_gen/sparse_bw_api_gen.py b/python/paddle/utils/code_gen/sparse_bw_api_gen.py
index ede4de2bdd67965d16823dfcd2e13d46fef64f2d..561e198a41b99c2362f5e14ac7fcd6da051c8875 100644
--- a/python/paddle/utils/code_gen/sparse_bw_api_gen.py
+++ b/python/paddle/utils/code_gen/sparse_bw_api_gen.py
@@ -111,7 +111,6 @@ def source_include(header_file_path):
 #include "paddle/phi/api/lib/kernel_dispatch.h"
 #include "paddle/phi/api/lib/sparse_api_custom_impl.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/declarations.h"
 """