[SYCLomatic] Fix the issue that group utils header file not inserted but required (#2776)

Signed-off-by: intwanghao <hao3.wang@intel.com>

Author: intwanghao

clang/lib/DPCT/AnalysisInfo.cpp

@@ -6113,9 +6113,14 @@ void KernelCallExpr::addAccessorDecl() {
     Tex.second->addDecl(OuterStmts.InitList, SubmitStmts.TextureList,
                         SubmitStmts.SamplerList, getQueueStr());
   }
-  for (auto &Tmp : VM.getTempStorageMap()) {
-    Tmp.second->addAccessorDecl(SubmitStmts.AccessorList,
-                                ExecutionConfig.LocalSize);
+  if (!VM.getTempStorageMap().empty()) {
+    DpctGlobalInfo::getInstance()
+        .insertFile(getFilePath())
+        ->insertHeader(HT_DPCT_GROUP_Utils, RT_ForSYCLMigration);
+    for (auto &Tmp : VM.getTempStorageMap()) {
+      Tmp.second->addAccessorDecl(SubmitStmts.AccessorList,
+                                  ExecutionConfig.LocalSize);
+    }
   }
 }
 void KernelCallExpr::buildInfo() {

clang/lib/DPCT/RulesLangLib/CUBAPIMigration.cpp

@@ -247,6 +247,9 @@ void CubMemberCallRule::runRule(
   if (const auto *BlockMC =
           getNodeAsType<CXXMemberCallExpr>(Result, "memberCall")) {
     EA.analyze(BlockMC);
+    emplaceTransformation(EA.getReplacement());
+    EA.applyAllSubExprRepl();
+
     StringRef Name = BlockMC->getMethodDecl()->getName();
     bool isBlockRadixSort = Name == "Sort" || Name == "SortDescending" ||
                             Name == "SortBlockedToStriped" ||
@@ -343,9 +346,9 @@ void CubMemberCallRule::runRule(
     }
   } else if (const auto *E2 = getNodeAsType<MemberExpr>(Result, "memberExpr")) {
     EA.analyze(E2);
+    emplaceTransformation(EA.getReplacement());
+    EA.applyAllSubExprRepl();
   }
-  emplaceTransformation(EA.getReplacement());
-  EA.applyAllSubExprRepl();
 }
 
 void CubIntrinsicRule::registerMatcher(ast_matchers::MatchFinder &MF) {

clang/test/dpct/cub/blocklevel/blockload.cu

@@ -6,7 +6,7 @@
 // RUN: %if build_lit %{icpx -c -fsycl %T/blocklevel/blockload/blockload.dp.cpp -o %T/blocklevel/blockload/blockload.dp.o %}
 
 #include <cub/cub.cuh>
-
+// CHECK: #include <dpct/group_utils.hpp>
 __global__ void DirectKernel(int *d_data, int valid_items, int default_value) {
   int thread_data[4] = {0};
 // CHECK:  dpct::group::load_direct_blocked(item_ct1, d_data, thread_data);