!2097 review gpu quant code

Merge pull request !2097 from chenzhongming/master

mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold2_gpu_kernel.h

@@ -39,12 +39,10 @@ class BatchNormFold2GpuKernel : public GpuKernel {
   ~BatchNormFold2GpuKernel() override { DestroyResource(); }
 
   const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
-
   const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
-
   const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
 
-  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
               const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
     if (is_null_input_) {
       return true;
@@ -111,10 +109,7 @@ class BatchNormFold2GpuKernel : public GpuKernel {
     input_size_list_.push_back(weight_size);      // running_std
     input_size_list_.push_back(weight_size);      // running_mean
     input_size_list_.push_back(sizeof(int32_t));  // global_step
-
     output_size_list_.push_back(input_size);
-
-    workspace_size_list_.push_back(sizeof(int32_t));
   }
 
  private:

mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold2_grad_gpu_kernel.h

@@ -39,9 +39,7 @@ class BatchNormFold2GradGpuKernel : public GpuKernel {
   ~BatchNormFold2GradGpuKernel() override { DestroyResource(); }
 
   const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
-
   const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
-
   const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
 
   bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,

mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_gpu_kernel.h

@@ -47,9 +47,7 @@ class BatchNormFoldGpuKernel : public GpuKernel {
   ~BatchNormFoldGpuKernel() override { DestroyResource(); }
 
   const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
-
   const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
-
   const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
 
   bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,

mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_grad_gpu_kernel.h

@@ -46,9 +46,8 @@ class BatchNormFoldGradGpuKernel : public GpuKernel {
   const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
   const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
 
-  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
               const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
-    (void)workspace;
     // 'd_batch_mean', 'd_batch_std', 'x', 'batch_mean', 'batch_std', 'current_step'
     T *d_batch_mean = GetDeviceAddress<T>(inputs, 0);
     T *d_batch_std = GetDeviceAddress<T>(inputs, 1);
@@ -139,11 +138,8 @@ class BatchNormFoldGradGpuKernel : public GpuKernel {
     input_size_list_.push_back(channel_size_);
     input_size_list_.push_back(channel_size_);
     input_size_list_.push_back(sizeof(int));
-
     // 'dx'
     output_size_list_.push_back(input_size_);
-
-    workspace_size_list_.push_back(workspace_size_);
   }
 
  private:

mindspore/ccsrc/kernel/gpu/quant/correction_mul_gpu_kernel.h

@@ -33,7 +33,8 @@ class CorrectionMulGpuKernel : public GpuKernel {
   const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
   const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
   const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
-  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
               const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
     auto *weight = GetDeviceAddress<T>(inputs, 0);
     auto *gamma = GetDeviceAddress<T>(inputs, 1);
@@ -74,10 +75,9 @@ class CorrectionMulGpuKernel : public GpuKernel {
     input_size_list_.push_back(input_size);   // weight
     input_size_list_.push_back(weight_size);  // gamma
     input_size_list_.push_back(weight_size);  // running_std
-    size_t workspace_size = 0;
     output_size_list_.push_back(input_size);
-    workspace_size_list_.push_back(workspace_size);
   }
+
   void InitResource() override {}
 
  private:

mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.cc

@@ -101,10 +101,9 @@ void FakeQuantGradGpuKernel::InitSizeLists() {
   input_size_list_.push_back(min_size_);    // min
   input_size_list_.push_back(max_size_);    // max
   output_size_list_.push_back(output_size_);
-  workspace_size_list_.push_back(workspace_size_);
 }
 
-bool FakeQuantGradGpuKernel::Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+bool FakeQuantGradGpuKernel::Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
                                     const std::vector<AddressPtr> &outputs, void *stream_ptr) {
   float *output = GetDeviceAddress<float>(outputs, 0);
   float *gradient = GetDeviceAddress<float>(inputs, 0);