follow comments

paddle/fluid/memory/detail/system_allocator.cc

@@ -131,7 +131,12 @@ void* CUDAPinnedAllocator::Alloc(size_t& index, size_t size) {
   size_t usable =
       paddle::platform::CUDAPinnedMaxAllocSize() - cuda_pinnd_alloc_size_;
 
-  if (size > usable) return nullptr;
+  if (size > usable) {
+    LOG(WARNING) << "Cannot malloc " << size / 1024.0 / 1024.0
+                 << " MB pinned memory."
+                 << ", available " << usable / 1024.0 / 1024.0 << " MB";
+    return nullptr;
+  }
 
   void* p;
   // PINNED memory is visible to all CUDA contexts.
@@ -141,6 +146,9 @@ void* CUDAPinnedAllocator::Alloc(size_t& index, size_t size) {
     index = 1;  // PINNED memory
     cuda_pinnd_alloc_size_ += size;
     return p;
+  } else {
+    LOG(WARNING) << "cudaMallocHost failed.";
+    return nullptr;
   }
 
   return nullptr;

paddle/fluid/memory/memory.cc

@@ -133,11 +133,10 @@ void* Alloc<platform::CUDAPinnedPlace>(platform::CUDAPinnedPlace place,
   auto* buddy_allocator = GetCUDAPinnedBuddyAllocator();
   void* ptr = buddy_allocator->Alloc(size);
 
-  //  if (ptr == nullptr) {
-  //    LOG(WARNING) << "Cannot allocate " << size << " bytes in CUDAPinnedPlace
-  //    "
-  //                 << ", available " << avail << " bytes"
-  //  }
+  if (ptr == nullptr) {
+    LOG(WARNING) << "cudaMallocHost Cannot allocate " << size
+                 << " bytes in CUDAPinnedPlace";
+  }
   return ptr;
 }
 

paddle/fluid/platform/cpu_info.cc

@@ -27,8 +27,9 @@ DEFINE_double(fraction_of_cpu_memory_to_use, 1,
               "Default use 100% of CPU memory for PaddlePaddle,"
               "reserve the rest for page tables, etc");
 
-DEFINE_double(fraction_of_cuda_pinned_memory_to_use, 0.5,
-              "Default use 100% of CPU memory for PaddlePaddle,"
+DEFINE_double(
+    fraction_of_cuda_pinned_memory_to_use, 0.5,
+    "Default use 50% of CPU memory as the pinned_memory for PaddlePaddle,"
     "reserve the rest for page tables, etc");
 
 namespace paddle {
@@ -78,7 +79,7 @@ size_t CUDAPinnedMinChunkSize() {
 }
 
 size_t CUDAPinnedMaxChunkSize() {
-  // Allow to allocate the maximum chunk size is roughly 0.39% of CUDA_PINNED
+  // Allow to allocate the maximum chunk size is roughly 1/256 of CUDA_PINNED
   // memory.
   return CUDAPinnedMaxAllocSize() / 256;
 }