[Phi] Move mul op kernel into phi

paddle/fluid/framework/ir/memory_optimize_pass/share_varinfo_into_cinn_pass_test.cc

@@ -24,7 +24,7 @@
 #include "paddle/fluid/framework/parallel_executor.h"
 #include "paddle/fluid/framework/program_desc.h"
 
-USE_OP(mul);
+USE_OP_ITSELF(mul);
 USE_OP(cinn_launch);
 USE_OP_ITSELF(elementwise_add);
 namespace paddle::framework {

paddle/fluid/framework/paddle2cinn/build_cinn_pass_test.cc

@@ -674,7 +674,7 @@ TEST(BuildCinnPassTest, NoNeedBufferInput) {
 }  // namespace paddle
 
 USE_PASS(build_cinn_pass);
-USE_OP(mul);
+USE_OP_ITSELF(mul);
 USE_OP_ITSELF(relu);
 USE_OP_ITSELF(elementwise_add);
 USE_OP_ITSELF(relu_grad);

paddle/fluid/framework/paddle2cinn/cinn_compiler_test.cc

@@ -300,6 +300,6 @@ TEST(CinnCompilerTest, Compile) {
 
 USE_PASS(build_cinn_pass);
 USE_PASS(graph_viz_pass);
-USE_OP(mul);
+USE_OP_ITSELF(mul);
 USE_OP_ITSELF(relu);
 USE_OP_ITSELF(elementwise_add);

paddle/fluid/imperative/tests/test_eager.cc

@@ -98,4 +98,4 @@ TEST(test_var_helper, eager_var_helper) {
 }  // namespace imperative
 }  // namespace paddle
 
-USE_OP(mul);
+USE_OP_ITSELF(mul);

paddle/fluid/imperative/tests/test_hooks.cc

@@ -28,6 +28,8 @@
 
 PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_with_flatten, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_with_flatten_grad, CPU, ALL_LAYOUT);
 
 namespace platform = paddle::platform;
 namespace framework = paddle::framework;
@@ -267,7 +269,7 @@ TEST(TestHooks, TestGradVarLeafBackwardHookWithSortedGradAccmulated) {
 }  // namespace imperative
 }  // namespace paddle
 
-USE_OP(mul);
-USE_OP(mul_grad);
+USE_OP_ITSELF(mul);
+USE_OP_ITSELF(mul_grad);
 USE_OP_ITSELF(elementwise_add);
 USE_OP_ITSELF(elementwise_add_grad);

paddle/fluid/imperative/tests/test_layer.cc

@@ -416,4 +416,4 @@ TEST(test_layer, test_eager) {
 }  // namespace imperative
 }  // namespace paddle
 
-USE_OP(mul);
+USE_OP_ITSELF(mul);

paddle/fluid/imperative/tests/test_tracer.cc

@@ -34,9 +34,13 @@ PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(add_grad, CPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(sum, CPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(sum_grad, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_with_flatten, CPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_with_flatten_grad, CPU, ALL_LAYOUT);
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 PD_DECLARE_KERNEL(add_grad, GPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(sum_grad, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_with_flatten, GPU, ALL_LAYOUT);
+PD_DECLARE_KERNEL(matmul_with_flatten_grad, GPU, ALL_LAYOUT);
 #endif
 
 namespace imperative = paddle::imperative;
@@ -598,8 +602,8 @@ TEST(test_tracer, eager_tracer) {
 }  // namespace imperative
 }  // namespace paddle
 
-USE_OP(mul);
-USE_OP(mul_grad);
+USE_OP_ITSELF(mul);
+USE_OP_ITSELF(mul_grad);
 USE_OP_ITSELF(reduce_sum);
 USE_OP_ITSELF(reduce_sum_grad);
 USE_OP_ITSELF(elementwise_add);

paddle/fluid/inference/tensorrt/convert/test_fc_op.cc

@@ -43,4 +43,4 @@ TEST(fc_op, test) {
 }  // namespace tensorrt
 }  // namespace inference
 }  // namespace paddle
-USE_OP(mul);
+USE_OP_ITSELF(mul);

paddle/fluid/inference/tensorrt/convert/test_mul_op.cc

@@ -46,4 +46,4 @@ TEST(MulOpConverter, main) {
 }  // namespace inference
 }  // namespace paddle
 
-USE_OP(mul);
+USE_OP_ITSELF(mul);

paddle/fluid/operators/mkldnn/mul_mkldnn_op.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include <string>
 
-#include "paddle/fluid/operators/mul_op.h"
+#include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/platform/mkldnn_reuse.h"
 
 namespace phi {
@@ -46,6 +46,9 @@ using dnnl::memory;
 using dnnl::prop_kind;
 using dnnl::stream;
 
+constexpr int kMULMKLDNNINT8 = 1;
+constexpr int kMULMKLDNNFP32 = 2;
+
 template <typename XT, typename YT, typename OT>
 class MulPrimitiveFactory {
  public:

paddle/fluid/operators/mul_op.cc

@@ -12,11 +12,11 @@ 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/mul_op.h"
 #include <memory>
 #include <string>
 #include <unordered_map>
 #include <vector>
+#include "paddle/fluid/framework/op_registry.h"
 #ifdef PADDLE_WITH_MKLDNN
 #include "paddle/fluid/platform/mkldnn_helper.h"
 #endif
@@ -27,6 +27,9 @@ namespace operators {
 using framework::OpKernelType;
 using framework::Tensor;
 
+constexpr int kMULMKLDNNINT8 = 1;
+constexpr int kMULMKLDNNFP32 = 2;
+
 class MulOp : public framework::OperatorWithKernel {
  public:
   using framework::OperatorWithKernel::OperatorWithKernel;
@@ -354,16 +357,3 @@ REGISTER_OPERATOR(mul_grad, ops::MulGradOp,
                   ops::MulDoubleGradMaker<paddle::imperative::OpBase>);
 
 REGISTER_OPERATOR(mul_grad_grad, ops::MulDoubleGradOp);
-
-REGISTER_OP_CPU_KERNEL(
-    mul, ops::MulKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::MulKernel<paddle::platform::CPUDeviceContext, double>);
-
-REGISTER_OP_CPU_KERNEL(
-    mul_grad, ops::MulGradKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::MulGradKernel<paddle::platform::CPUDeviceContext, double>);
-
-REGISTER_OP_CPU_KERNEL(
-    mul_grad_grad,
-    ops::MulDoubleGradKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::MulDoubleGradKernel<paddle::platform::CPUDeviceContext, double>);