Polish some PE code details (#21274)

* polish code details, test=develop

* futher polish hint msg, test=develop

paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_optimizer_op_pass.cc

@@ -55,7 +55,7 @@ void FuseOptimizerOpPass::ApplyImpl(ir::Graph *graph) const {
   VLOG(6) << "Find " << fuse_op_type << " operators : " << opt_ops_num
           << ", and " << opt_nodes.size() << " for dense gradients.";
 
-  if (opt_nodes.size() == 0) return;
+  if (opt_nodes.size() <= 1) return;
   if (result.Has(details::kFusedOptType)) {
     auto &opt_type = result.Get<details::FusedOptType>(details::kFusedOptType);
     VLOG(6) << "Currently only support fusing one type of optimizer op, "

paddle/fluid/framework/parallel_executor.cc

@@ -421,10 +421,9 @@ ParallelExecutor::ParallelExecutor(const std::vector<platform::Place> &places,
 #endif
 
   LOG(INFO) << string::Sprintf(
-      "The number of %s, which is used in ParallelExecutor, is %lu. And "
-      "the Program will be copied %lu copies",
-      (member_->use_cuda_ ? "CUDAPlace" : "CPUPlace"), places.size(),
-      places.size());
+      "The Program will be executed on %s using ParallelExecutor, %lu "
+      "cards are used, so %lu programs are executed in parallel.",
+      (member_->use_cuda_ ? "CUDA" : "CPU"), places.size(), places.size());
 
   // Step 1. Bcast the bcast_vars to devs.
   // Create local scopes

python/paddle/fluid/tests/unittests/test_buffer_shared_memory_reuse_pass_and_fuse_optimization_op_pass.py

@@ -28,7 +28,7 @@ class CUDAInplaceTestWithFuseOptimizationOps(InplaceTestBase):
         self.check_single_card_fetch_var()
 
 
-class CPUAInplaceTestWithFuseOptimizationOps(InplaceTestBase):
+class CPUInplaceTestWithFuseOptimizationOps(InplaceTestBase):
     def initParameter(self):
         self.use_cuda = False
         self.fuse_all_optimizer_ops = True