!3658 set output format to nhwc4

Merge pull request !3658 from wandongdong/master_ori

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

@@ -82,10 +82,13 @@ int ArithmeticOpenCLKernel::Init() {
   ocl_runtime->LoadSource(program_name, source);
   ocl_runtime->BuildKernel(kernel_, program_name, kernel_name, build_options);
 #endif
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 
 int ArithmeticOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   uint32_t element_num = outputs_[0]->ElementsC4Num();
   auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
   std::vector<size_t> global = {element_num};

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

@@ -35,7 +35,7 @@ int ConcatOpenCLKernel::Init() {
   }
 
   auto param = reinterpret_cast<ConcatParameter *>(this->opParameter);
-  MS_LOG(INFO) << "concat at axis=:  " << param->axis_;
+  MS_LOG(DEBUG) << "concat at axis=:  " << param->axis_;
   if (param->axis_ != 0 && param->axis_ != 3) {
     MS_LOG(ERROR) << "only support axis=0 or axis=3";
   }
@@ -55,6 +55,8 @@ int ConcatOpenCLKernel::Init() {
   ocl_runtime->LoadSource(program_name, source);
   ocl_runtime->BuildKernel(kernel_, program_name, kernel_name, build_options);
 #endif
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 
@@ -88,6 +90,7 @@ int ConcatOpenCLKernel::Run_axis0() {
 }
 
 int ConcatOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   auto param = reinterpret_cast<ConcatParameter *>(this->opParameter);
   if (param->axis_ == 0) {
     return Run_axis0();

mindspore/lite/src/runtime/kernel/opencl/kernel/conv2d_transpose.cc

@@ -69,6 +69,8 @@ int Conv2dTransposeOpenCLKernel::Init() {
   PadWeight();
   allocator->UnmapBuffer(padWeight_);
   allocator->UnmapBuffer(bias_);
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 
@@ -120,6 +122,7 @@ void Conv2dTransposeOpenCLKernel::PadWeight() {
 }
 
 int Conv2dTransposeOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   std::vector<int> shapex = inputs_[0]->shape();
   int n = shapex[0];
   if (n > 1) {

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

@@ -38,12 +38,8 @@ int ConvolutionOpenCLKernel::Init() {
     MS_LOG(ERROR) << "ConvolutionOpenCLKernel only support Batch=1!";
   }
 
-  auto io_NHWC = inputs_[0]->GetFormat() == schema::Format_NHWC && outputs_[0]->GetFormat() == schema::Format_NHWC;
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
-  auto io_NHWC4 = inputs_[0]->GetFormat() == schema::Format_NHWC4 && outputs_[0]->GetFormat() == schema::Format_NHWC4;
+  io_dataformat_ = outputs_[0]->GetFormat();
-  if (!io_NHWC && !io_NHWC4) {
-    MS_LOG(ERROR) << "input and output data_format is invalid!";
-  }
-  io_dataformat_ = inputs_[0]->GetFormat();
 
   if (inputs_[1]->GetFormat() != schema::Format_KHWC) {
     MS_LOG(ERROR) << "weight data_format is invalid!";
@@ -62,6 +58,7 @@ int ConvolutionOpenCLKernel::Init() {
 #endif
 
   this->InitBuffer();
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 int ConvolutionOpenCLKernel::InitBuffer() {
@@ -123,7 +120,7 @@ int ConvolutionOpenCLKernel::InitBuffer() {
 int ConvolutionOpenCLKernel::ReSize() { return 0; }
 
 int ConvolutionOpenCLKernel::Run() {
-  MS_LOG(INFO) << "ConvolutionOpenCLKernel::Run()";
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
 
   auto param = reinterpret_cast<ConvParameter *>(opParameter);

mindspore/lite/src/runtime/kernel/opencl/kernel/depthwise_conv2d.cc

@@ -58,6 +58,7 @@ int DepthwiseConv2dOpenCLKernel::Init() {
   ocl_runtime->BuildKernel(kernel_, program_name, kernel_name, build_options);
 #endif
   this->InitBuffer();
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 int DepthwiseConv2dOpenCLKernel::InitBuffer() {
@@ -100,6 +101,7 @@ int DepthwiseConv2dOpenCLKernel::ReSize() {
 }
 
 int DepthwiseConv2dOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   auto parameter = reinterpret_cast<ConvParameter*>(opParameter);
   auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
   size_t CO4 = UP_DIV(outputs_[0]->Channel(), C4NUM);

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

@@ -66,6 +66,8 @@ int MatMulOpenCLKernel::Init() {
   if (hasBias_) {
     allocator->UnmapBuffer(bias_);
   }
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 
@@ -100,6 +102,7 @@ void MatMulOpenCLKernel::PadWeight() {
 }
 
 int MatMulOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   std::vector<int> shapex = inputs_[0]->shape();
   int n = shapex[0];
   if (n > 1) {

mindspore/lite/src/runtime/kernel/opencl/kernel/pooling2d.cc

@@ -59,8 +59,6 @@ int PoolingOpenCLKernel::Init() {
     MS_LOG(ERROR) << "Init `Pooling2d` kernel failed!";
     return RET_INVALID_OP_NAME;
   }
-  auto in_format = inputs_[0]->GetFormat();
-  outputs_[0]->SetFormat(in_format);
   auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
 
 #ifdef PROGRAM_WITH_IL
@@ -70,6 +68,8 @@ int PoolingOpenCLKernel::Init() {
   ocl_runtime->LoadSource(program_name, source);
   ocl_runtime->BuildKernel(kernel_, program_name, kernel_name, build_options);
 #endif
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return RET_OK;
 }
 
@@ -85,6 +85,7 @@ int PoolingOpenCLKernel::InitBuffer() { return 0; }
 int PoolingOpenCLKernel::ReSize() { return 0; }
 
 int PoolingOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
 
   // attribute

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

@@ -46,6 +46,8 @@ int SoftmaxOpenCLKernel::Init() {
   ocl_runtime->LoadSource(program_name, source);
   ocl_runtime->BuildKernel(kernel_, program_name, kernel_name, build_options);
 #endif
+  outputs_[0]->SetFormat(schema::Format_NHWC4);
+  MS_LOG(DEBUG) << kernel_name << " Init Done!";
   return 0;
 }
 
@@ -53,6 +55,7 @@ int SoftmaxOpenCLKernel::InitBuffer() { return 0; }
 int SoftmaxOpenCLKernel::ReSize() { return 0; }
 
 int SoftmaxOpenCLKernel::Run() {
+  MS_LOG(DEBUG) << this->Name() << " Running!";
   auto ocl_runtime = lite::opencl::OpenCLRuntime::GetInstance();
   auto allocator = ocl_runtime->GetAllocator();