fix cuda117 compile error and rm useless code; test=develop (#59915)

Author: danleifeng

paddle/fluid/framework/fleet/heter_ps/graph_gpu_wrapper.h

@@ -22,10 +22,12 @@
 #include "paddle/fluid/framework/fleet/heter_ps/gpu_graph_node.h"
 
 #ifdef PADDLE_WITH_HETERPS
+#if defined(__NVCC__) || defined(__HIPCC__)
 #include <thrust/device_ptr.h>
 #include <thrust/execution_policy.h>
 #include <thrust/random.h>
 #include <thrust/shuffle.h>
+#endif
 #include "paddle/fluid/framework/fleet/heter_ps/mem_pool.h"
 #include "paddle/fluid/platform/device/gpu/gpu_info.h"
 #include "paddle/fluid/platform/dynload/nccl.h"

paddle/fluid/framework/tensor_util.cc

@@ -307,8 +307,6 @@ void TensorCopySync(const phi::DenseTensor& src,
     return;
   }
 
-  VLOG(0) << "TensorCopySync " << src.dims() << " from " << src.place()
-          << " to " << dst_place;
   src.check_memory_size();
   dst->Resize(src.dims());
   dst->set_layout(src.layout());

python/paddle/distributed/fleet/base/strategy_compiler.py

@@ -94,7 +94,6 @@ def maximum_path_len_algo(optimizer_list):
     edge, indegree = create_graph(optimizer_list)
     topo_sort(edge, indegree)
     max_path = floyd(edge)
-    print("max_path=", max_path)
     candidate = []
     for idx in max_path:
         candidate.append(optimizer_list[idx])

python/paddle/incubate/operators/unzip.py

@@ -25,6 +25,7 @@ def unzip(input, lod, len):
 
     Args:
         input (Variable): The zipped input
+        len(int): The second dim length of unzipped output.
         lod (Variable): The original lod of unzipped input, 1-D LodTensor with shape[K].
 
     Returns: