gpu minddata 0921

cmake/options.cmake

@@ -121,8 +121,7 @@ if(ENABLE_TESTCASES OR (NOT ENABLE_D))
     add_compile_definitions(NO_DLIB=1)
 endif()
 
-if(NOT (ENABLE_TESTCASES OR ENABLE_TEST) AND NOT (CMAKE_SYSTEM_NAME MATCHES "Windows" OR
-        CMAKE_SYSTEM_NAME MATCHES "Darwin") AND (ENABLE_D OR ENABLE_GPU OR ENABLE_CPU))
+if(NOT (ENABLE_TESTCASES OR ENABLE_TEST) AND (ENABLE_D OR ENABLE_GPU OR ENABLE_CPU))
     add_compile_definitions(WITH_BACKEND)
 endif()
 

mindspore/ccsrc/backend/common/session/session_basic.cc

@@ -48,7 +48,7 @@
 #include "utils/trace_base.h"
 #include "include/common/utils/parallel_context.h"
 #include "kernel/oplib/oplib.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/ps_cache/ps_data/ps_data_prefetch.h"
 #include "ps/constants.h"
 #include "ps/util.h"

mindspore/ccsrc/common/mem_reuse/mem_reuse_checker.h

@@ -29,7 +29,7 @@
 namespace mindspore {
 namespace memreuse {
 constexpr auto kSplitC = '/';
-class MemReuseChecker {
+class BACKEND_EXPORT MemReuseChecker {
  public:
   static MemReuseChecker &GetInstance();
   MemReuseChecker(const MemReuseChecker &) = delete;

mindspore/ccsrc/distributed/init.h

@@ -18,7 +18,7 @@
 #define MINDSPORE_CCSRC_DISTRIBUTED_INIT_H_
 
 #include "distributed/collective/collective_manager.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "distributed/cluster/cluster_context.h"
 #else
 #include "distributed/cluster/dummy_cluster_context.h"

mindspore/ccsrc/frontend/parallel/graph_util/graph_splitter.cc

@@ -28,7 +28,7 @@
 #include "include/common/utils/anfalgo.h"
 #include "include/common/debug/draw.h"
 #include "include/common/utils/parallel_context.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/ps_context.h"
 #endif
 
@@ -336,7 +336,7 @@ distributed::DistExecutionMode GenerateStrategy() {
   distributed::DistExecutionMode strategy;
   bool enable_ps = false;
   bool enable_embedding_cache = false;
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   enable_ps = ps::PSContext::instance()->is_ps_mode();
   enable_embedding_cache = ps::PSContext::instance()->cache_enable();
 #endif

mindspore/ccsrc/frontend/parallel/graph_util/graph_splitter.h

@@ -29,7 +29,7 @@
 #include "include/common/utils/utils.h"
 #include "ir/func_graph.h"
 #include "distributed/constants.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "distributed/cluster/cluster_context.h"
 #else
 #include "distributed/cluster/dummy_cluster_context.h"

mindspore/ccsrc/frontend/parallel/ops_info/gather_info.cc

@@ -26,9 +26,8 @@
 #include "frontend/parallel/dynamic_creator.h"
 #include "frontend/parallel/graph_util/generate_graph.h"
 #include "include/common/utils/parallel_context.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/ps_cache/ps_data/ps_data_prefetch.h"
-#include "utils/ms_context.h"
 #include "ps/ps_context.h"
 #include "distributed/embedding_cache/embedding_cache_utils.h"
 #endif
@@ -163,7 +162,7 @@ Status GatherInfo::GetAttrs() {
   if (std::find(inputs_shape_[1].begin(), inputs_shape_[1].end(), -1) != inputs_shape_[1].end()) {
     dynamic_shape_indices_ = true;
   }
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::PsDataPrefetch::GetInstance().cache_enable()) {
     dynamic_shape_indices_ = true;
   }
@@ -757,7 +756,7 @@ Status GatherInfo::InferBias() {
         rank = rank % (params_strategy[0] * params_strategy[1]);
       }
     }
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
     if (ps::PsDataPrefetch::GetInstance().cache_enable()) {
       if (ps::PSContext::instance()->enable_distributed_mindrt()) {
         bias_ = static_cast<int64_t>(embedding_cache_table_manager.cache_indices_lower_bound());

mindspore/ccsrc/frontend/parallel/ops_info/unique_info.cc

@@ -29,7 +29,7 @@
 #include "frontend/parallel/strategy.h"
 #include "include/common/utils/parallel_context.h"
 #include "frontend/parallel/tensor_layout/tensor_redistribution.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/ps_cache/ps_data/ps_data_prefetch.h"
 #include "ps/ps_context.h"
 #include "distributed/embedding_cache/embedding_cache_utils.h"
@@ -102,7 +102,7 @@ std::vector<StrategyPtr> UniqueInfo::GenerateOpStrategies(int64_t stage_id) {
 
   return sp_vector;
 }
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 Status UniqueInfo::ComputeReplaceGraph(const CNodePtr &cnode) {
   GenerateGraph gen_g = GenerateGraph(attrs_);
   if (gen_g.Init(cnode) != SUCCESS) {
@@ -142,7 +142,7 @@ Status UniqueInfo::ComputeReplaceGraph(const CNodePtr &cnode) {
 }
 #endif
 ReplaceGraphPtr UniqueInfo::replace_graph(const CNodePtr &cnode) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::PsDataPrefetch::GetInstance().cache_enable()) {
     auto inputs = cnode->inputs();
     if (inputs.empty()) {

mindspore/ccsrc/frontend/parallel/step_auto_parallel.cc

@@ -48,7 +48,7 @@
 #include "ir/anf.h"
 #include "ir/param_info.h"
 #include "ir/tensor.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/util.h"
 #endif
 
@@ -71,7 +71,7 @@ void SearchParallelStrategy(const std::string &strategy_search_mode, const FuncG
 }
 
 bool StepAutoParallel(const FuncGraphPtr &root, const opt::OptimizerPtr &) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::Util::IsRoleOfPServer() || ps::Util::IsRoleOfScheduler()) {
     return false;
   }

mindspore/ccsrc/minddata/dataset/engine/execution_tree.cc

@@ -21,6 +21,7 @@
 #include "minddata/dataset/engine/datasetops/data_queue_op.h"
 #ifdef WITH_BACKEND
 #include "mindspore/core/utils/numa_interface.h"
+#include "utils/ms_context.h"
 #endif
 #include "minddata/dataset/util/task_manager.h"
 #include "minddata/dataset/util/service.h"

mindspore/ccsrc/pipeline/jit/action.cc

@@ -53,7 +53,7 @@
 #include "backend/graph_compiler/transform.h"
 #include "load_mindir/infer_mindir.h"
 #include "debug/data_dump/dump_json_parser.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/scheduler.h"
 #include "distributed/cluster/cluster_context.h"
 #endif
@@ -127,7 +127,7 @@ void DisableMindRT(const ResourcePtr &resource) {
   if (context_ptr->get_param<bool>(MS_CTX_ENABLE_MINDRT) == false) {
     return;
   }
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::PSContext::instance()->cache_enable()) {
     return;
   }
@@ -863,7 +863,7 @@ bool OptInlineAction(const ResourcePtr &resource) {
 bool GeOptimizeAction(const ResourcePtr &resource) { return OptimizeAction(resource, kGePasses); }
 
 bool VmOptimizeAction(const ResourcePtr &resource) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::PSContext::instance()->is_ps_mode()) {
     (void)kVmPasses.emplace_back(PassItem("server_communication_op_fusion", ps::Util::FuseServerCommOps));
   }
@@ -1121,7 +1121,7 @@ void SetRunMode(const FuncGraphPtr &func_graph, compile::Backend *backend_ptr) {
     return;
   }
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::PSContext::instance()->cache_enable()) {
     set_ctx(true, false, false);
     return;
@@ -1182,7 +1182,7 @@ void SetRunMode(const ResourcePtr &resource) {
   auto enable_hccl = context_ptr->get_param<bool>(MS_CTX_ENABLE_HCCL);
   // After the distributed interface on D is unified, the following flag and judgment will be removed.
   bool enbale_distributed_mindrt = false;
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   enbale_distributed_mindrt = ps::PSContext::instance()->enable_distributed_mindrt();
 #endif
   if (!is_task_sink && mode == kGraphMode && enable_hccl && !common::UseMPI() && !enbale_distributed_mindrt) {
@@ -1278,7 +1278,7 @@ bool ExecuteAction(const ResourcePtr &resource) {
   return true;
 }
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 bool StartPSSchedulerAction(const ResourcePtr &) {
   if (distributed::cluster::ClusterContext::instance()->initialized()) {
     MS_LOG(INFO) << "This node is scheduler. Start wait for finalizing.";
@@ -1574,7 +1574,7 @@ std::vector<ActionItem> VmPipeline(const ResourcePtr &resource) {
     (void)actions.emplace_back(std::make_pair("validate", ValidateAction));
   }
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   (void)actions.emplace_back(std::make_pair("distribtued_split", DistributedSplitAction));
   if (ps::PSContext::instance()->is_worker()) {
     if (distributed::cluster::ClusterContext::instance()->initialized()) {
@@ -1611,7 +1611,7 @@ std::vector<ActionItem> MindIRPipeline() {
   return actions;
 }
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 std::vector<ActionItem> PSchedulerPipeline(const ResourcePtr &resource) {
   if (resource->EnableCompileCache() && resource->func_graph() != nullptr) {
     return {std::make_pair("scheduler", StartPSSchedulerAction)};

mindspore/ccsrc/pipeline/jit/action.h

@@ -37,13 +37,17 @@ bool GeOptimizeAction(const ResourcePtr &resource);
 bool VmOptimizeAction(const ResourcePtr &resource);
 bool TaskEmitAction(const ResourcePtr &resource);
 bool ExecuteAction(const ResourcePtr &resource);
+#if defined(__linux__) && defined(WITH_BACKEND)
 bool StartPSSchedulerAction(const ResourcePtr &resource);
 bool DistributedSplitAction(const ResourcePtr &resource);
+#endif
 
 std::vector<ActionItem> GePipeline();
 std::vector<ActionItem> VmPipeline(const ResourcePtr &resource);
 std::vector<ActionItem> MindIRPipeline();
+#if defined(__linux__) && defined(WITH_BACKEND)
 std::vector<ActionItem> PSchedulerPipeline(const ResourcePtr &resource);
+#endif
 abstract::AnalysisResult AbstractAnalyze(const ResourcePtr &resource, const FuncGraphPtr &func_graph,
                                          const abstract::AbstractBasePtrList &args_abs, bool clear = false);
 FuncGraphPtr ProgramSpecialize(const ResourcePtr &resource, const FuncGraphPtr &func_graph,

mindspore/ccsrc/pipeline/jit/compile_cache_manager.cc

@@ -30,7 +30,7 @@
 #include "frontend/parallel/step_parallel.h"
 #include "mindspore/core/utils/file_utils.h"
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/core/node.h"
 #include "distributed/cluster/cluster_context.h"
 #endif
@@ -69,7 +69,7 @@ std::string GetCompileCacheDir() {
 }
 
 std::string GetRole() {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (distributed::cluster::ClusterContext::instance()->initialized()) {
     auto node = distributed::cluster::ClusterContext::instance()->node();
     MS_EXCEPTION_IF_NULL(node);

mindspore/ccsrc/pipeline/jit/pipeline.cc

@@ -67,7 +67,7 @@
 #include "runtime/device/stream_synchronizer.h"
 #include "distributed/collective/collective_manager.h"
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/constants.h"
 #include "ps/util.h"
 #include "ps/ps_cache/ps_data/ps_data_prefetch.h"
@@ -748,7 +748,7 @@ std::vector<ActionItem> GetPipeline(const ResourcePtr &resource, const std::stri
   auto ms_context = MsContext::GetInstance();
   MS_EXCEPTION_IF_NULL(ms_context);
   std::string backend = ms_context->backend_policy();
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (distributed::cluster::ClusterContext::instance()->initialized()) {
     auto node = distributed::cluster::ClusterContext::instance()->node();
     MS_EXCEPTION_IF_NULL(node);
@@ -1448,7 +1448,7 @@ bool InitExecDataset(const std::string &queue_name, int64_t iter_num, int64_t ba
 bool InitExecDatasetVm(const std::string &queue_name, int64_t size, int64_t batch_size,
                        const std::vector<TypePtr> &types, const std::vector<std::vector<int64_t>> &shapes,
                        const std::vector<int64_t> &input_indexes, bool need_run) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if ((ps::PSContext::instance()->is_ps_mode()) && (!ps::PSContext::instance()->is_worker())) {
     return true;
   }
@@ -1503,7 +1503,7 @@ bool InitExecDatasetVm(const std::string &queue_name, int64_t size, int64_t batc
   MS_EXCEPTION_IF_NULL(backend);
   // The data set graph compiling and running of mindRT.
   if (MsContext::GetInstance()->get_param<bool>(MS_CTX_ENABLE_MINDRT)) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
     if (ps::PSContext::instance()->is_worker() && ps::PSContext::instance()->cache_enable()) {
       ps::PsDataPrefetch::GetInstance().CreateDataChannel(queue_name, LongToSize(size));
     }
@@ -1529,7 +1529,7 @@ bool InitExecDatasetVm(const std::string &queue_name, int64_t size, int64_t batc
   auto runner = convert_fn(segment, "");
   ConfigManager::GetInstance().set_iter_num(queue_name, size);
   // PS cache does not support loop sink.
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ps::PSContext::instance()->is_worker() && ps::PsDataPrefetch::GetInstance().cache_enable()) {
     ps::PsDataPrefetch::GetInstance().CreateDataChannel(queue_name, LongToSize(size));
     ConfigManager::GetInstance().set_iter_num(queue_name, 1);
@@ -1898,7 +1898,7 @@ bool PyIsCipherFile(const std::string &file_path) { return mindspore::IsCipherFi
 
 void FinalizeCluster() {
   MS_LOG(INFO) << "Start finalize the cluster instance.";
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (distributed::cluster::ClusterContext::instance()->initialized()) {
     runtime::EmbeddingCacheScheduler::GetInstance().Finalize();
     (void)distributed::cluster::ClusterContext::instance()->Finalize(UINT32_MAX);

mindspore/ccsrc/plugin/device/cpu/hal/hardware/cpu_device_context.cc

@@ -47,7 +47,7 @@
 #include "backend/common/session/anf_runtime_algorithm.h"
 #include "include/common/utils/anfalgo.h"
 #include "plugin/device/cpu/hal/profiler/cpu_profiling.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "plugin/device/cpu/hal/hardware/ms_collective_comm_lib.h"
 #endif
 #ifndef ENABLE_SECURITY
@@ -407,7 +407,7 @@ bool CPUDeviceResManager::LoadCollectiveCommLib() {
     collective_comm_lib_ = instance_func();
     MS_EXCEPTION_IF_NULL(collective_comm_lib_);
   } else {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
     collective_comm_lib_ = &MsCollectiveCommLib::GetInstance();
     MS_EXCEPTION_IF_NULL(collective_comm_lib_);
 #endif

mindspore/ccsrc/plugin/device/cpu/kernel/allreduce_cpu_kernel.cc

@@ -20,13 +20,13 @@
 #include <functional>
 #include <memory>
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "plugin/device/cpu/hal/hardware/ms_collective_comm_lib.h"
 #endif
 
 namespace mindspore {
 namespace kernel {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 using device::CollectiveOpReduceType::Reduce_Sum;
 using device::cpu::kMCCLGlobalGroupName;
 using device::cpu::MsCollectiveCommLib;
@@ -37,7 +37,7 @@ constexpr char kSupportedReduceOp[] = "sum";
 }  // namespace
 
 void AllReduceCPUKernelMod::InitKernel(const CNodePtr &kernel_node) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   MS_EXCEPTION_IF_NULL(kernel_node);
   kernel_name_ = common::AnfAlgo::GetCNodeName(kernel_node);
   auto kernel_attr = GetKernelAttrFromNode(kernel_node);
@@ -67,7 +67,7 @@ std::vector<KernelAttr> AllReduceCPUKernelMod::GetOpSupport() {
 bool AllReduceCPUKernelMod::Launch(const std::vector<kernel::AddressPtr> &inputs,
                                    const std::vector<kernel::AddressPtr> &,
                                    const std::vector<kernel::AddressPtr> &outputs) {
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (inputs.empty() || outputs.empty()) {
     MS_LOG(EXCEPTION) << kernel_name_ << " has at least one input and one output, but got 0.";
   }

mindspore/ccsrc/plugin/device/gpu/hal/device/gpu_data_queue.cc

@@ -50,6 +50,10 @@ DataQueueStatus GpuDataQueueDynamic::Push(std::vector<DataQueueItem> data) {
     return DataQueueStatus::SUCCESS;
   }
 
+  if (IsFull()) {
+    return DataQueueStatus::TIMEOUT;
+  }
+
   for (size_t i = 0; i < data.size(); i++) {
     auto &item = data[i];
     if (item.data_ptr == nullptr) {
@@ -196,6 +200,9 @@ DataQueueStatus GpuDataQueue::Push(std::vector<DataQueueItem> data) {
   if (data.empty()) {
     return DataQueueStatus::SUCCESS;
   }
+  if (IsFull()) {
+    return DataQueueStatus::TIMEOUT;
+  }
   size_t data_len =
     std::accumulate(data.begin(), data.end(), 0, [](size_t accum, const auto &it) { return accum + it.data_len; });
   void *device_addr = nullptr;

mindspore/ccsrc/runtime/graph_scheduler/actor/actor_set.h

@@ -17,7 +17,7 @@
 #ifndef MINDSPORE_CCSRC_RUNTIME_FRAMEWORK_ACTOR_SET_H_
 #define MINDSPORE_CCSRC_RUNTIME_FRAMEWORK_ACTOR_SET_H_
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #define ENABLE_RPC_ACTOR
 #endif
 

mindspore/ccsrc/runtime/graph_scheduler/actor/data_prepare_actor.cc

@@ -26,7 +26,7 @@
 #include "utils/log_adapter.h"
 #include "include/common/utils/convert_utils.h"
 #include "distributed/recovery/recovery_context.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "runtime/graph_scheduler/embedding_cache_scheduler.h"
 #endif
 
@@ -926,7 +926,7 @@ void DataPrepareActor::PreprocessBeforePrepareData() const {
   // Embedding Cache mode needs to record the number of global steps executed by the compute graph.
   // The first step compute graph needs to wait for the Embedding cache prefetch cache to warm up to prevent the
   // GetNext operator from timing out in the compute graph.
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   EmbeddingCacheScheduler::GetInstance().IncreaseGraphStep(GetAID());
 #endif
 }

mindspore/ccsrc/runtime/graph_scheduler/actor/loop_count_actor.cc

@@ -27,7 +27,7 @@
 #include "runtime/device/stream_synchronizer.h"
 #include "distributed/recovery/recovery_context.h"
 #include "distributed/collective/collective_manager.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "runtime/graph_scheduler/rpc_node_scheduler.h"
 #endif
 
@@ -107,7 +107,7 @@ void LoopCountActor::SendOutput(OpContext<DeviceTensor> *const context) {
     ActorDispatcher::Send(entrance_aid, &EntranceActor::ClearDataOnStepEnd, from_aid, context);
   }
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   // Update rpc actors' status.
   RpcActorStatusUpdater::GetInstance().UpdateRpcActorStatus();
 #endif

mindspore/ccsrc/runtime/graph_scheduler/graph_compiler.cc

@@ -44,7 +44,7 @@
 #ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
 #endif
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "ps/ps_context.h"
 #include "runtime/graph_scheduler/embedding_cache_scheduler.h"
 #endif
@@ -374,7 +374,7 @@ GraphId GraphCompiler::CompileWholeGraphForGraphRunMode(const FuncGraphPtr &func
   // set executing sink true in graph mode
   root_graph->set_run_mode(device::RunMode::kGraphMode);
   root_graph->set_is_loop_count_sink(true);
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   // Embedding cache need global step of compute graph, can not enable loop sink, move loop control to loop count actor.
   if (ps::PSContext::instance()->cache_enable()) {
     root_graph->set_is_loop_count_sink(false);
@@ -473,7 +473,7 @@ GraphId GraphCompiler::CompileGraphImpl(const KernelGraphPtr &graph, const Devic
   DumpJsonParser::GetInstance().UpdateNeedDumpKernels(*graph.get());
 #endif
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   // Set device address for embedding cache parameter, only enable when enable embedding cache mode.
   EmbeddingCacheScheduler::GetInstance().SetEmbedCachedParamAddress(device_context, graph);
 #endif

mindspore/ccsrc/runtime/graph_scheduler/graph_scheduler.cc

@@ -54,7 +54,7 @@
 #include "include/common/debug/common.h"
 #include "distributed/recovery/recovery_context.h"
 #include "distributed/collective/collective_manager.h"
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 #include "distributed/cluster/cluster_context.h"
 #else
 #include "distributed/cluster/dummy_cluster_context.h"
@@ -195,7 +195,7 @@ void IntHandler(int, siginfo_t *, void *) {
 }
 #endif
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
 void SendFinishTransform(const std::string &actor_set_name) {
   auto node = ClusterContext::instance()->node();
   MS_EXCEPTION_IF_NULL(node);
@@ -517,13 +517,13 @@ ActorSet *GraphScheduler::Transform(const GraphCompilerInfo &graph_compiler_info
   Optimize(actor_set);
   MS_LOG(INFO) << "Graph(" << graph_compiler_info.name_ << ") transforms actor end.";
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   if (ClusterContext::instance()->initialized() && RecoveryContext::GetInstance()->enable_recovery()) {
     SendFinishTransform(graph_compiler_info.name_);
   }
 #endif
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   // Save data channel for this actor set.
   MS_EXCEPTION_IF_NULL(actor_set->data_prepare_actor_);
   EmbeddingCacheScheduler::GetInstance().SetDataSetChannel(actor_set->data_prepare_actor_->GetAID(),
@@ -637,7 +637,7 @@ void GraphScheduler::Run(ActorSet *const actor_set, const std::vector<std::vecto
   const size_t kSecondsToMilliseconds = 1000;
   SetActorExecutionStrategy(actor_set, strategy, (end_time - start_time) * kSecondsToMilliseconds);
 
-#ifdef WITH_BACKEND
+#if defined(__linux__) && defined(WITH_BACKEND)
   DoDisasterRecovery(actor_set->name_);
 #endif
 }