Fix bug in CUDA

Author: reyoung

paddle/fluid/operators/random_crop_op.h

@@ -67,7 +67,7 @@ HOSTDEVICE inline void StridedMemcpy(const T* x, const size_t* x_dims, T* out,
     }
   } else {
     x += offset_i * x_stride;
-    for (size_t j = 0; j < x_dim_i; ++j) {
+    for (size_t j = 0; j < out_dim_i; ++j) {
       StridedMemcpy<T>(x, x_dims, out, out_dims, i + 1, rank, x_stride,
                        out_stride, offsets);
       x += x_stride;
@@ -86,8 +86,6 @@ struct RandomCropFunctor {
   int rank_;
   int64_t seed_;
 
-  size_t prod_x_dims_;
-  size_t prod_out_dims_;
   size_t prod_batchsize_dims_;
   size_t prod_x_ins_dims_;
   size_t prod_out_ins_dims_;
@@ -118,8 +116,6 @@ struct RandomCropFunctor {
         prod_out_ins_dims_ *= out_dim_i;
       }
     }
-    prod_x_dims_ = prod_batchsize_dims_ * prod_x_ins_dims_;
-    prod_out_dims_ = prod_batchsize_dims_ * prod_out_ins_dims_;
   }
 
   HOSTDEVICE void operator()(size_t ins_idx) {
@@ -146,7 +142,17 @@ template <typename DeviceContext, typename T>
 class RandomCropKernel : public framework::OpKernel<T> {
  public:
   virtual void Compute(const framework::ExecutionContext& ctx) const {
-    int64_t seed = *ctx.Input<framework::LoDTensor>("Seed")->data<int64_t>();
+    auto& seed_tensor = detail::Ref(ctx.Input<framework::LoDTensor>("Seed"));
+    int64_t seed = 0;
+    if (platform::is_cpu_place(seed_tensor.place())) {
+      seed = *seed_tensor.data<int64_t>();
+    } else {
+      LOG(WARNING) << "It is slow to place seed in GPU memory. Please verify "
+                      "your program";
+      framework::LoDTensor cpu_seed;
+      framework::TensorCopySync(seed_tensor, platform::CPUPlace(), &cpu_seed);
+      seed = *cpu_seed.data<int64_t>();
+    }
     auto shape = ctx.Attr<std::vector<int>>("shape");
     auto& x = detail::Ref(ctx.Input<framework::LoDTensor>("X"));
     auto& out = detail::Ref(ctx.Output<framework::LoDTensor>("Out"));