op cache supports un-persistable attributes (#43221)

paddle/fluid/framework/operator.cc

@@ -1529,8 +1529,20 @@ void OperatorWithKernel::RunImpl(const Scope& scope,
       // Do data transform before building KernelContext
       // TODO(zhiqiu): support TransferInplaceVarsBack
       PreparePhiData(exec_scope, *pt_kernel_, *kernel_signature_, runtime_ctx);
-      BuildPhiKernelContext(*runtime_ctx, dev_ctx, &pt_kernel_context);
-      (*pt_kernel_)(&pt_kernel_context);
+      if (enable_cache_runtime_context_ && !need_prepare_phi_data_ &&
+          !need_prepare_data_) {
+        impl_ =
+            new CacheImpl(new phi::KernelContext(),
+                          new RuntimeInferShapeContext(*this, *runtime_ctx));
+        BuildPhiKernelContext(*runtime_ctx, dev_ctx, impl_->getKernelContext());
+        (*pt_kernel_)(impl_->getKernelContext());
+      } else {
+        phi::KernelContext pt_kernel_context;
+        // Do data transform before building KernelContext
+        // TODO(zhiqiu): support TransferInplaceVarsBack
+        BuildPhiKernelContext(*runtime_ctx, dev_ctx, &pt_kernel_context);
+        (*pt_kernel_)(&pt_kernel_context);
+      }
     } else {
       (*kernel_func_)(
           ExecutionContext(*this, exec_scope, *dev_ctx, *runtime_ctx));
@@ -2386,7 +2398,6 @@ void OperatorWithKernel::BuildPhiKernelContext(
     // calcute the start and end index of the input tensors
     size_t start_idx =
         (i == 0 ? 0 : pt_kernel_context->InputRangeAt(i - 1).second);
-
     // deal with optional here
     if ((it == ctx.inputs.end() || it->second.size() == 0) &&
         (input_defs[i].type_index ==
@@ -2400,6 +2411,7 @@ void OperatorWithKernel::BuildPhiKernelContext(
       auto end_idx = start_idx + 1;
       pt_kernel_context->AssignInputRange(std::make_pair(start_idx, end_idx),
                                           i);
+
       continue;
     }
     auto ins_vector = it->second;
@@ -2414,6 +2426,7 @@ void OperatorWithKernel::BuildPhiKernelContext(
         tensor_in = &(var->Get<phi::SelectedRows>());
         pt_kernel_context->EmplaceBackInputWithoutSetRange(tensor_in);
       } else if (var->IsType<framework::LoDTensorArray>()) {
+        need_prepare_phi_data_ = true;
         paddle::small_vector<const phi::TensorBase*> tensor_vector;
         auto& tensor_array = var->Get<framework::LoDTensorArray>();
         for (auto& t : tensor_array) {
@@ -2514,6 +2527,7 @@ void OperatorWithKernel::BuildPhiKernelContext(
                   attr_names[i]));
           }
         } else {  // scalar is in the input
+          need_prepare_phi_data_ = true;
           auto& ins_vector = ctx.inputs.at(attr_names[i]);
           pt_kernel_context->EmplaceBackAttr(std::move(
               experimental::MakePhiScalarFromVar(*ins_vector.front())));
@@ -2545,6 +2559,7 @@ void OperatorWithKernel::BuildPhiKernelContext(
                   attr_names[i]));
           }
         } else {  // shape is in the input
+          need_prepare_phi_data_ = true;
           auto& ins_vector = ctx.inputs.at(attr_names[i]);
           if (ins_vector.size() == 1) {  // ShapeTensor
             pt_kernel_context->EmplaceBackAttr(std::move(