!10326 add memory pool total size limits

From: @hangangqiang
Reviewed-by: 
Signed-off-by:

mindspore/lite/src/executor.cc

@@ -49,11 +49,12 @@ int Executor::Run(std::vector<Tensor *> &in_tensors, std::vector<Tensor *> &out_
     MS_LOG(ERROR) << "CheckInputs failed";
     return ret;
   }
-  MS_ASSERT(std::all_of(kernels.begin(), kernels.end(), [](kernel::LiteKernel *kernel) {
-    return std::all_of(kernel->in_tensors().begin(), kernel->in_tensors().end(), [](Tensor *in_tensor) {
-      return in_tensor->IsConst() || in_tensor->IsGraphInput() || in_tensor->ref_count() == 0;
-    });
-  }));
+  // clear ref_count
+  for (auto *kernel : kernels) {
+    for (auto *tensor : kernel->in_tensors()) {
+      tensor->set_ref_count(0);
+    }
+  }
   std::queue<kernel::LiteKernel *> kernel_queue;
   for (auto kernel : kernels) {
     if (kernel->IsReady(kernel->in_tensors())) {

mindspore/lite/src/lite_kernel.cc

@@ -115,7 +115,7 @@ int LiteKernel::PreProcess() {
 
   for (auto *output : this->out_tensors()) {
     MS_ASSERT(output != nullptr);
-    if (output->ElementsNum() >= MAX_MALLOC_SIZE / static_cast<int>(sizeof(int64_t))) {
+    if (output->ElementsNum() >= lite::MAX_MALLOC_SIZE / static_cast<int>(sizeof(int64_t))) {
       MS_LOG(ERROR) << "The size of output tensor is too big";
       return RET_ERROR;
     }

mindspore/lite/src/lite_session.cc

@@ -251,6 +251,9 @@ void LiteSession::InitGraphInputMap(const lite::Model *model) {
       }
       auto tensor_name = in_node->name_ + std::to_string(i);
       this->input_map_[tensor_name] = in_tensor;
+      if (!in_tensor->tensor_name().empty()) {
+        this->input_map_[in_tensor->tensor_name()] = in_tensor;
+      }
     }
   }
 }

mindspore/lite/src/runtime/allocator.cc

@@ -49,6 +49,10 @@ void *DefaultAllocator::Malloc(size_t size) {
     MS_LOG(ERROR) << "MallocData out of max_size, size: " << size;
     return nullptr;
   }
+  if (this->GetTotalSize() >= MAX_THREAD_POOL_SIZE) {
+    MS_LOG(ERROR) << "Memory pool is exhausted";
+    return nullptr;
+  }
   Lock();
   auto iter = freeList_.lower_bound(size);
   if (iter != freeList_.end() && (iter->second->size >= size) && (iter->second->size < (size << shiftFactor_))) {

mindspore/lite/src/runtime/allocator.h

@@ -39,7 +39,6 @@ class Allocator {
   virtual void Free(void *ptr) = 0;
   virtual void SetContext(const AllocatorContext &ctx) {}
   virtual size_t GetTotalSize() { return 0; }
-  virtual void Clear() {}
   static std::shared_ptr<Allocator> Create();
   virtual void *Prepare(void *ptr) { return ptr; }
   std::string name;
@@ -53,7 +52,7 @@ class DefaultAllocator : public Allocator {
   void *Malloc(size_t size) override;
   void Free(void *ptr) override;
   size_t GetTotalSize() override;
-  void Clear() override;
+  void Clear();
 
  private:
   void Lock();
@@ -72,7 +71,8 @@ class DefaultAllocator : public Allocator {
   bool lockFlag_ = false;
 };
 
-#define MAX_MALLOC_SIZE (2000 * 1024 * 1024)
+constexpr int64_t MAX_MALLOC_SIZE = static_cast<size_t>(2000) * 1024 * 1024;
+constexpr int64_t MAX_THREAD_POOL_SIZE = static_cast<size_t>(3000) * 1024 * 1024;
 
 }  // namespace mindspore::lite
 

mindspore/lite/src/runtime/opencl/opencl_allocator.h

@@ -42,7 +42,7 @@ class OpenCLAllocator : public Allocator {
   void Free(void *ptr) override;
   size_t GetTotalSize() override;
 
-  void Clear() override;
+  void Clear();
   void *GetImage(void *host_ptr);
   void *GetBuffer(void *host_ptr);
   void *MapBuffer(void *host_ptr, int flags, void *command_queue = nullptr, bool sync = true);