[Phi] Remove needless deps in unittests (#40256)

* remove needless deps in unittests

* add gpu marco

* fix other unittests

* fix kernel name error

* fix test_prepare_op

* fix failed dygraph unittests

* fix gpu failed tests

* fix cinn test failed

* fix cinn test failed

* fix dropout tests

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 {
 

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
 

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
 

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());
 

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 {
 

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 {
 

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) {

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) {

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) {

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(

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) {

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(

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(

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) {

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 {

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;

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;

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;

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;

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;

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 {

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 {
 

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 {
 

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,

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;

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

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;

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 {
 

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 {

paddle/phi/tests/api/CMakeLists.txt

 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})

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 {
 

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 {
 

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 {
 

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 {

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 {
 

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 {
 

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 {
 

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 {
 

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 {
 

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 {
 

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 {
 

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 {

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 {
 

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 {
 

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 {
 

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 {
 

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,

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());

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 {
 

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 {
 

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 {
 

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"
 """
 
 

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"
 """
 
 

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"
 """