!5529 add biasadd support in arithmetic for opencl

Merge pull request !5529 from liuchao/master

mindspore/lite/src/runtime/kernel/opencl/kernel/arithmetic.cc

@@ -30,6 +30,14 @@ using mindspore::lite::KernelRegistrar;
 
 namespace mindspore::kernel {
 
+ArithmeticOpenCLKernel::~ArithmeticOpenCLKernel() {
+  if (weight_ptr_ != nullptr) {
+    auto allocator = runtime_->GetAllocator();
+    allocator->Free(weight_ptr_);
+    weight_ptr_ = nullptr;
+  }
+}
+
 std::vector<size_t> ArithmeticOpenCLKernel::InitGlobalSize() const {
   const size_t global_x = out_tensors_[0]->Width();
   const size_t global_y = out_tensors_[0]->Height();
@@ -39,10 +47,18 @@ std::vector<size_t> ArithmeticOpenCLKernel::InitGlobalSize() const {
 }
 
 void ArithmeticOpenCLKernel::Image2dGetWorkGroupSize() {
-  size_t H = out_tensors_[0]->Batch() * out_tensors_[0]->Height();
-  size_t W = out_tensors_[0]->Width() * UP_DIV(out_tensors_[0]->Channel(), C4NUM);
   local_size_ = {16, 16};
-  global_size_ = {W, H};
+  if (out_tensors_[0]->GetFormat() == schema::Format_NHWC4) {
+    size_t H = out_tensors_[0]->Batch() * out_tensors_[0]->Height();
+    size_t W = out_tensors_[0]->Width() * UP_DIV(out_tensors_[0]->Channel(), C4NUM);
+    global_size_ = {W, H};
+  } else if (out_tensors_[0]->GetFormat() == schema::Format_NC4) {
+    size_t H = out_tensors_[0]->Batch();
+    size_t W = UP_DIV(out_tensors_[0]->Channel(), C4NUM);
+    global_size_ = {W, H};
+  } else {
+    MS_LOG(ERROR) << "Unspport data format " << out_tensors_[0]->GetFormat();
+  }
 }
 
 void ArithmeticOpenCLKernel::BufferGetWorkGroupSize() {
@@ -51,16 +67,16 @@ void ArithmeticOpenCLKernel::BufferGetWorkGroupSize() {
 }
 
 int ArithmeticOpenCLKernel::GetImageSize(size_t idx, std::vector<size_t> *img_size) {
-  size_t CO4 = UP_DIV(out_tensors_[0]->Channel(), C4NUM);
-  int H = out_tensors_[0]->Batch() * out_tensors_[0]->Height();
-  int W = out_tensors_[0]->Width() * CO4;
   size_t im_dst_x, im_dst_y;
-  if (in_tensors_[0]->GetFormat() == schema::Format_NHWC4) {
-    im_dst_x = W;
-    im_dst_y = H;
+  if (out_tensors_[0]->GetFormat() == schema::Format_NHWC4) {
+    im_dst_x = out_tensors_[0]->Width() * UP_DIV(out_tensors_[0]->Channel(), C4NUM);
+    im_dst_y = out_tensors_[0]->Batch() * out_tensors_[0]->Height();
+  } else if (out_tensors_[0]->GetFormat() == schema::Format_NC4) {
+    im_dst_y = out_tensors_[0]->Batch();
+    im_dst_x = UP_DIV(out_tensors_[0]->Channel(), C4NUM);
   } else {
-    im_dst_y = out_tensors_[0]->Batch() * out_tensors_[0]->Height() * CO4;
-    im_dst_x = out_tensors_[0]->Width();
+    MS_LOG(ERROR) << "Unspport data format " << out_tensors_[0]->GetFormat();
+    return RET_ERROR;
   }
 #ifdef ENABLE_FP16
   size_t img_dtype = CL_HALF_FLOAT;
@@ -73,11 +89,26 @@ int ArithmeticOpenCLKernel::GetImageSize(size_t idx, std::vector<size_t> *img_si
   return RET_OK;
 }
 
+int ArithmeticOpenCLKernel::InitBuffer() {
+  const ArithmeticParameter *arithmetic_parameter = reinterpret_cast<const ArithmeticParameter *>(op_parameter_);
+  if (!arithmetic_parameter->broadcasting_) {
+    if (in_tensors_[1]->TensorType() == schema::NodeType_ValueNode && in_tensors_[1]->Data() != nullptr) {
+      auto allocatdor = runtime_->GetAllocator();
+      std::vector<size_t> img_size;
+      GetImageSize(0, &img_size);
+      weight_ptr_ = allocatdor->CreateImageFromHost(in_tensors_[1]->Data(), in_tensors_[1]->ElementsNum(), img_size);
+      return RET_OK;
+    }
+  }
+  return RET_OK;
+}
 int ArithmeticOpenCLKernel::Init() {
   runtime_ = lite::opencl::OpenCLRuntime::GetInstance();
   std::string kernel_name;
 
-  if (in_tensors_[1]->TensorType() == schema::NodeType_ValueNode && in_tensors_[1]->Data() != nullptr) {
+  const ArithmeticParameter *arithmetic_parameter = reinterpret_cast<const ArithmeticParameter *>(op_parameter_);
+  if (arithmetic_parameter->broadcasting_ && in_tensors_[1]->TensorType() == schema::NodeType_ValueNode &&
+      in_tensors_[1]->Data() != nullptr) {
     element_flag_ = false;
     kernel_name = "BoardcastArith";
   } else {
@@ -103,7 +134,7 @@ int ArithmeticOpenCLKernel::Init() {
 
   lite::STATUS error_code = RET_OK;
 #ifdef PROGRAM_WITH_IL
-  kernel_ = ocl_runtime->GetKernelFromBinary(kernel_name);
+  kernel_ = runtime_->GetKernelFromBinary(kernel_name);
 #else
   if (out_mem_type_ == OpenCLMemType::IMG) {
     kernel_name += "_IMG";
@@ -119,23 +150,31 @@ int ArithmeticOpenCLKernel::Init() {
   if (error_code != RET_OK) {
     return error_code;
   }
+
+  auto format = schema::Format_NHWC4;
+  if (arithmetic_parameter->ndim_ == 2) {
+    format = schema::Format_NC4;
+  }
   in_ori_format_ = in_tensors_[0]->GetFormat();
-  in_tensors_[0]->SetFormat(schema::Format_NHWC4);
   out_ori_format_ = out_tensors_[0]->GetFormat();
-  out_tensors_[0]->SetFormat(schema::Format_NHWC4);
+  in_tensors_[0]->SetFormat(format);
+  if (element_flag_ && in_tensors_[1]->TensorType() != schema::NodeType_ValueNode) {
+    in_tensors_[1]->SetFormat(format);
+  }
+  out_tensors_[0]->SetFormat(format);
   Image2dGetWorkGroupSize();
+  InitBuffer();
   return RET_OK;
 }
 
 int ArithmeticOpenCLKernel::Run() {
   MS_LOG(DEBUG) << this->name() << " Running!";
-  auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
 
   int arg_idx = 0;
-
-  ocl_runtime->SetKernelArg(kernel_, arg_idx++, in_tensors_[0]->Data());
+  runtime_->SetKernelArg(kernel_, arg_idx++, in_tensors_[0]->Data());
   if (element_flag_) {
-    runtime_->SetKernelArg(kernel_, arg_idx++, in_tensors_[1]->Data());
+    void *weight = weight_ptr_ == nullptr ? in_tensors_[1]->Data() : weight_ptr_;
+    runtime_->SetKernelArg(kernel_, arg_idx++, weight);
   } else {
     float value = static_cast<float *>(in_tensors_[1]->Data())[0];
     switch (op_parameter_->type_) {
@@ -155,23 +194,47 @@ int ArithmeticOpenCLKernel::Run() {
         MS_LOG(ERROR) << "Error Operator type " << op_parameter_->type_;
         break;
     }
-    ocl_runtime->SetKernelArg(kernel_, arg_idx++, weight_);
-    ocl_runtime->SetKernelArg(kernel_, arg_idx++, bias_);
+    runtime_->SetKernelArg(kernel_, arg_idx++, weight_);
+    runtime_->SetKernelArg(kernel_, arg_idx++, bias_);
+  }
+  runtime_->SetKernelArg(kernel_, arg_idx++, out_tensors_[0]->Data());
+
+  int H = 0;
+  int W = 0;
+  if (out_tensors_[0]->GetFormat() == schema::Format_NHWC4) {
+    H = out_tensors_[0]->Batch() * out_tensors_[0]->Height();
+    W = out_tensors_[0]->Width() * UP_DIV(out_tensors_[0]->Channel(), C4NUM);
+  } else if (out_tensors_[0]->GetFormat() == schema::Format_NC4) {
+    H = out_tensors_[0]->Batch();
+    W = UP_DIV(out_tensors_[0]->Channel(), C4NUM);
+  } else {
+    MS_LOG(ERROR) << "Error output type " << out_tensors_[0]->GetFormat();
+    return RET_ERROR;
   }
-  ocl_runtime->SetKernelArg(kernel_, arg_idx++, out_tensors_[0]->Data());
-  int H = out_tensors_[0]->Batch() * out_tensors_[0]->Height();
-  int W = out_tensors_[0]->Width() * UP_DIV(out_tensors_[0]->Channel(), C4NUM);
   cl_int2 output_shape{W, H};
-  ocl_runtime->SetKernelArg(kernel_, arg_idx++, output_shape);
-  ocl_runtime->RunKernel(kernel_, global_size_, local_size_, nullptr);
+  runtime_->SetKernelArg(kernel_, arg_idx++, output_shape);
+  runtime_->RunKernel(kernel_, global_size_, local_size_, nullptr);
   return RET_OK;
 }
 
+kernel::LiteKernel *OpenCLBiasAddKernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                               const std::vector<lite::tensor::Tensor *> &outputs,
+                                               OpParameter *opParameter, const lite::Context *ctx,
+                                               const kernel::KernelKey &desc, const lite::PrimitiveC *primitive);
+
 kernel::LiteKernel *OpenCLArithmeticKernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
                                                   const std::vector<lite::tensor::Tensor *> &outputs,
                                                   OpParameter *opParameter, const lite::Context *ctx,
                                                   const kernel::KernelKey &desc,
                                                   const mindspore::lite::PrimitiveC *primitive) {
+  const ArithmeticParameter *arithmetic_parameter = reinterpret_cast<const ArithmeticParameter *>(opParameter);
+  if (arithmetic_parameter->broadcasting_) {
+    for (size_t i = 0; i < arithmetic_parameter->ndim_; i++) {
+      if (arithmetic_parameter->in_shape1_[i] != 0 && arithmetic_parameter->in_shape1_[i] != 1) {
+        return OpenCLBiasAddKernelCreator(inputs, outputs, opParameter, ctx, desc, primitive);
+      }
+    }
+  }
   auto *kernel =
     new (std::nothrow) ArithmeticOpenCLKernel(reinterpret_cast<OpParameter *>(opParameter), inputs, outputs, ctx);
   if (kernel == nullptr) {

mindspore/lite/src/runtime/kernel/opencl/kernel/arithmetic.h

@@ -29,7 +29,7 @@ class ArithmeticOpenCLKernel : public OpenCLKernel {
   explicit ArithmeticOpenCLKernel(OpParameter *parameter, const std::vector<lite::tensor::Tensor *> &inputs,
                                   const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx)
       : OpenCLKernel(parameter, inputs, outputs) {}
-  ~ArithmeticOpenCLKernel() override{};
+  ~ArithmeticOpenCLKernel() override;
 
   int Init() override;
   int Run() override;
@@ -39,12 +39,14 @@ class ArithmeticOpenCLKernel : public OpenCLKernel {
   std::vector<size_t> InitGlobalSize() const;
   void Image2dGetWorkGroupSize();
   void BufferGetWorkGroupSize();
+  int InitBuffer();
 
   cl::Kernel kernel_;
   lite::opencl::OpenCLRuntime *runtime_;
   bool element_flag_{true};
   float weight_{1.f};
   float bias_{.0f};
+  void *weight_ptr_{nullptr};
 
   std::vector<size_t> local_size_;
   std::vector<size_t> global_size_;