Fix CUDA Graph H2D bug by restore host memory (#37774)

* fix CUDA Graph H2D bug again

* fix no return bug

paddle/fluid/operators/math/concat_and_split.cu

@@ -287,13 +287,11 @@ class ConcatFunctor<platform::CUDADeviceContext, T> {
     const T** dev_ins_data = nullptr;
     if (!has_same_shape || in_num < 2 || in_num > 4) {
       tmp_dev_ins_data = memory::Alloc(context, in_num * sizeof(T*));
-      {
-        platform::SkipCUDAGraphCaptureGuard guard;
+      auto* restored =
+          platform::RestoreHostMemIfCapturingCUDAGraph(inputs_data, in_num);
       memory::Copy(BOOST_GET_CONST(platform::CUDAPlace, context.GetPlace()),
-                     tmp_dev_ins_data->ptr(), platform::CPUPlace(),
-                     static_cast<void*>(inputs_data), in_num * sizeof(T*),
-                     context.stream());
-      }
+                   tmp_dev_ins_data->ptr(), platform::CPUPlace(), restored,
+                   in_num * sizeof(T*), context.stream());
       dev_ins_data = reinterpret_cast<const T**>(tmp_dev_ins_data->ptr());
     }
 
@@ -317,13 +315,12 @@ class ConcatFunctor<platform::CUDADeviceContext, T> {
     } else {
       auto tmp_dev_ins_col_data =
           memory::Alloc(context, inputs_col_num * sizeof(int64_t));
-      {
-        platform::SkipCUDAGraphCaptureGuard guard;
+
+      auto* restored = platform::RestoreHostMemIfCapturingCUDAGraph(
+          inputs_col, inputs_col_num);
       memory::Copy(BOOST_GET_CONST(platform::CUDAPlace, context.GetPlace()),
-                     tmp_dev_ins_col_data->ptr(), platform::CPUPlace(),
-                     static_cast<void*>(inputs_col),
+                   tmp_dev_ins_col_data->ptr(), platform::CPUPlace(), restored,
                    inputs_col_num * sizeof(int64_t), context.stream());
-      }
       int64_t* dev_ins_col_data =
           static_cast<int64_t*>(tmp_dev_ins_col_data->ptr());
 
@@ -422,13 +419,11 @@ class SplitFunctor<platform::CUDADeviceContext, T> {
     T** dev_out_gpu_data = nullptr;
     if (!has_same_shape || o_num < 2 || o_num > 4) {
       tmp_dev_outs_data = memory::Alloc(context, o_num * sizeof(T*));
-      {
-        platform::SkipCUDAGraphCaptureGuard guard;
+      auto* restored =
+          platform::RestoreHostMemIfCapturingCUDAGraph(outputs_data, o_num);
       memory::Copy(BOOST_GET_CONST(platform::CUDAPlace, context.GetPlace()),
-                     tmp_dev_outs_data->ptr(), platform::CPUPlace(),
-                     reinterpret_cast<void*>(outputs_data), o_num * sizeof(T*),
-                     context.stream());
-      }
+                   tmp_dev_outs_data->ptr(), platform::CPUPlace(), restored,
+                   o_num * sizeof(T*), context.stream());
       dev_out_gpu_data = reinterpret_cast<T**>(tmp_dev_outs_data->ptr());
     }
 
@@ -452,13 +447,11 @@ class SplitFunctor<platform::CUDADeviceContext, T> {
     } else {
       auto tmp_dev_ins_col_data =
           memory::Alloc(context, outputs_cols_num * sizeof(int64_t));
-      {
-        platform::SkipCUDAGraphCaptureGuard guard;
+      auto* restored = platform::RestoreHostMemIfCapturingCUDAGraph(
+          outputs_cols, outputs_cols_num);
       memory::Copy(BOOST_GET_CONST(platform::CUDAPlace, context.GetPlace()),
-                     tmp_dev_ins_col_data->ptr(), platform::CPUPlace(),
-                     reinterpret_cast<void*>(outputs_cols),
+                   tmp_dev_ins_col_data->ptr(), platform::CPUPlace(), restored,
                    outputs_cols_num * sizeof(int64_t), context.stream());
-      }
       int64_t* dev_outs_col_data =
           reinterpret_cast<int64_t*>(tmp_dev_ins_col_data->ptr());
 

paddle/fluid/platform/cuda_graph_with_memory_pool.h

@@ -60,6 +60,23 @@ inline void AddResetCallbackIfCapturingCUDAGraph(Callback &&callback) {
   callback();
 }
 
+template <typename T>
+inline T *RestoreHostMemIfCapturingCUDAGraph(T *host_mem, size_t size) {
+  static_assert(std::is_trivial<T>::value, "T must be trivial type");
+  static_assert(!std::is_same<T, void>::value, "T cannot be void");
+#ifdef PADDLE_WITH_CUDA
+  if (UNLIKELY(IsCUDAGraphCapturing())) {
+    size_t nbytes = size * sizeof(T);
+    void *new_host_mem = new uint8_t[nbytes];
+    std::memcpy(new_host_mem, host_mem, nbytes);
+    AddResetCallbackIfCapturingCUDAGraph(
+        [new_host_mem] { delete[] reinterpret_cast<uint8_t *>(new_host_mem); });
+    return reinterpret_cast<T *>(new_host_mem);
+  }
+#endif
+  return host_mem;
+}
+
 class SkipCUDAGraphCaptureGuard {
   DISABLE_COPY_AND_ASSIGN(SkipCUDAGraphCaptureGuard);