bugfix:SigmoidCrossEntropyWithLogitsGrad need multiply dout

mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sigmoid_cross_entropy_with_logits_grad_impl.cu

@@ -18,24 +18,24 @@
 
 template <typename T, typename S>
 __global__ void SigmoidCrossEntropyWithLogitsGradKernel(const size_t size, const T *logits, const S *labels,
-                                                        T *outputs) {
+                                                        const T *dout_addr, T *outputs) {
   for (size_t i = blockIdx.x * blockDim.x + threadIdx.x; i < size; i += gridDim.x * blockDim.x) {
     if (logits[i] >= 0) {
-      outputs[i] = 1. / (1. + exp(-logits[i])) - labels[i];
+      outputs[i] = (1. / (1. + exp(-logits[i])) - labels[i]) * dout_addr[i];
     } else {
       const T exp_val = exp(logits[i]);
-      outputs[i] = exp_val / (1. + exp_val) - labels[i];
+      outputs[i] = (exp_val / (1. + exp_val) - labels[i]) * dout_addr[i];
     }
   }
 }
 
 template <typename T, typename S>
-void SigmoidCrossEntropyWithLogitsGrad(const size_t size, const T *logits, const S *labels, T *outputs,
-                                       cudaStream_t cuda_stream) {
+void SigmoidCrossEntropyWithLogitsGrad(const size_t size, const T *logits, const S *labels, const T *dout_addr,
+                                       T *outputs, cudaStream_t cuda_stream) {
   SigmoidCrossEntropyWithLogitsGradKernel<<<GET_BLOCKS(size), GET_THREADS, 0, cuda_stream>>>(size, logits, labels,
-                                                                                             outputs);
+                                                                                             dout_addr, outputs);
 }
 
 template void SigmoidCrossEntropyWithLogitsGrad<float, float>(const size_t size, const float *logits,
-                                                              const float *labels, float *outputs,
-                                                              cudaStream_t cuda_stream);
+                                                              const float *labels, const float *dout_addr,
+                                                              float *outputs, cudaStream_t cuda_stream);

mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/sigmoid_cross_entropy_with_logits_grad_impl.cuh

@@ -19,7 +19,7 @@
 
 #include "runtime/device/gpu/cuda_common.h"
 template <typename T, typename S>
-void SigmoidCrossEntropyWithLogitsGrad(const size_t size, const T *logits, const S *labels, T *outputs,
-                                       cudaStream_t cuda_stream);
+void SigmoidCrossEntropyWithLogitsGrad(const size_t size, const T *logits, const S *labels, const T *dout_addr,
+                                       T *outputs, cudaStream_t cuda_stream);
 
 #endif  // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMP_SIGMOID_CROSS_ENTROPY_WITH_LOGITS_GRAD_IMPL_H_

mindspore/ccsrc/backend/kernel_compiler/gpu/nn/sigmoid_cross_entropy_with_logits_grad_gpu_kernel.h

@@ -38,9 +38,10 @@ class SigmoidCrossEntropyWithLogitsGradGpuKernel : public GpuKernel {
               const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
     T *logits_addr = GetDeviceAddress<T>(inputs, 0);
     S *labels_addr = GetDeviceAddress<S>(inputs, 1);
+    T *dout_addr = GetDeviceAddress<T>(inputs, 2);
     T *outputs_addr = GetDeviceAddress<T>(outputs, 0);
 
-    SigmoidCrossEntropyWithLogitsGrad(inputs[0]->size / sizeof(T), logits_addr, labels_addr, outputs_addr,
+    SigmoidCrossEntropyWithLogitsGrad(inputs[0]->size / sizeof(T), logits_addr, labels_addr, dout_addr, outputs_addr,
                                       reinterpret_cast<cudaStream_t>(stream_ptr));
     return true;
   }
@@ -78,6 +79,7 @@ class SigmoidCrossEntropyWithLogitsGradGpuKernel : public GpuKernel {
   void InitSizeLists() override {
     input_size_list_.push_back(logits_size_);
     input_size_list_.push_back(labels_size_);
+    input_size_list_.push_back(logits_size_);
     output_size_list_.push_back(outputs_size_);
   }