merge main into amd-staging (#540)

Author: z1-cciauto

clang/lib/Analysis/LifetimeSafety/Facts.cpp

@@ -68,8 +68,7 @@ llvm::StringMap<ProgramPoint> FactManager::getTestPoints() const {
     for (const Fact *F : BlockFacts) {
       if (const auto *TPF = F->getAs<TestPointFact>()) {
         StringRef PointName = TPF->getAnnotation();
-        assert(AnnotationToPointMap.find(PointName) ==
-                   AnnotationToPointMap.end() &&
+        assert(!AnnotationToPointMap.contains(PointName) &&
                "more than one test points with the same name");
         AnnotationToPointMap[PointName] = F;
       }

compiler-rt/test/lit.common.cfg.py

@@ -113,18 +113,16 @@ def push_dynamic_library_lookup_path(config, new_path):
         config.environment[dynamic_library_lookup_var] = new_ld_library_path_64
 
 
-# TODO: Consolidate the logic for turning on the internal shell by default for all LLVM test suites.
-# See https://github.com/llvm/llvm-project/issues/106636 for more details.
-#
 # Choose between lit's internal shell pipeline runner and a real shell.  If
 # LIT_USE_INTERNAL_SHELL is in the environment, we use that as an override.
 use_lit_shell = os.environ.get("LIT_USE_INTERNAL_SHELL")
 if use_lit_shell:
     # 0 is external, "" is default, and everything else is internal.
     execute_external = use_lit_shell == "0"
 else:
-    # Otherwise we default to internal everywhere.
-    execute_external = False
+    # Otherwise we default to internal on Windows and external elsewhere, as
+    # bash on Windows is usually very slow.
+    execute_external = not sys.platform in ["win32"]
 
 # Allow expanding substitutions that are based on other substitutions
 config.recursiveExpansionLimit = 10

llvm/include/llvm/CodeGen/TargetInstrInfo.h

@@ -436,7 +436,10 @@ class LLVM_ABI TargetInstrInfo : public MCInstrInfo {
   /// MachineSink determines on its own whether the instruction is safe to sink;
   /// this gives the target a hook to override the default behavior with regards
   /// to which instructions should be sunk.
+  ///
+  /// shouldPostRASink() is used by PostRAMachineSink.
   virtual bool shouldSink(const MachineInstr &MI) const { return true; }
+  virtual bool shouldPostRASink(const MachineInstr &MI) const { return true; }
 
   /// Return false if the instruction should not be hoisted by MachineLICM.
   ///

llvm/lib/CodeGen/MachineSink.cpp

@@ -2287,6 +2287,10 @@ bool PostRAMachineSinkingImpl::tryToSinkCopy(MachineBasicBlock &CurBB,
       continue;
     }
 
+    // Don't postRASink instructions that the target prefers not to sink.
+    if (!TII->shouldPostRASink(MI))
+      continue;
+
     if (MI.isDebugOrPseudoInstr())
       continue;
 

llvm/lib/ExecutionEngine/Orc/TargetProcess/SimpleExecutorMemoryManager.cpp

@@ -36,8 +36,6 @@ Expected<ExecutorAddr> SimpleExecutorMemoryManager::reserve(uint64_t Size) {
 
 Expected<ExecutorAddr>
 SimpleExecutorMemoryManager::initialize(tpctypes::FinalizeRequest &FR) {
-  std::vector<shared::WrapperFunctionCall> DeallocationActions;
-
   if (FR.Segments.empty()) {
     if (FR.Actions.empty())
       return make_error<StringError>("Finalization request is empty",

llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp

@@ -5421,8 +5421,6 @@ combineUnpackingMovIntoLoad(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) {
   if (!NVPTX::isPackedVectorTy(ElementVT) || ElementVT == MVT::v4i8)
     return SDValue();
 
-  SmallVector<SDNode *> DeadCopyToRegs;
-
   // Check whether all outputs are either used by an extractelt or are
   // glue/chain nodes
   if (!all_of(N->uses(), [&](SDUse &U) {

llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp

@@ -303,8 +303,6 @@ PreservedAnalyses LoopIdiomRecognizePass::run(Loop &L, LoopAnalysisManager &AM,
   // but ORE cannot be preserved (see comment before the pass definition).
   OptimizationRemarkEmitter ORE(L.getHeader()->getParent());
 
-  std::optional<PolynomialInfo> HR;
-
   LoopIdiomRecognize LIR(&AR.AA, &AR.DT, &AR.LI, &AR.SE, &AR.TLI, &AR.TTI,
                          AR.MSSA, DL, ORE);
   if (!LIR.runOnLoop(&L))

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -21222,7 +21222,6 @@ BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP,
     }
     ScheduledBundlesList.pop_back();
     SmallVector<ScheduleData *> ControlDependentMembers;
-    SmallPtrSet<Instruction *, 4> Visited;
     for (Value *V : VL) {
       if (S.isNonSchedulable(V))
         continue;

llvm/lib/Transforms/Vectorize/VPlan.cpp

@@ -52,10 +52,6 @@
 using namespace llvm;
 using namespace llvm::VPlanPatternMatch;
 
-namespace llvm {
-extern cl::opt<bool> EnableVPlanNativePath;
-}
-
 /// @{
 /// Metadata attribute names
 const char LLVMLoopVectorizeFollowupAll[] = "llvm.loop.vectorize.followup_all";

offload/include/omptarget.h

@@ -107,10 +107,6 @@ enum TargetAllocTy : int32_t {
   TARGET_ALLOC_DEFAULT,
 };
 
-inline KernelArgsTy CTorDTorKernelArgs = {
-    1,       0, nullptr,   nullptr,   nullptr,   nullptr, nullptr,
-    nullptr, 0, {0, 0, 0}, {1, 0, 0}, {1, 0, 0}, 0};
-
 struct DeviceTy;
 
 /// The libomptarget wrapper around a __tgt_async_info object directly