Merged main:b590ba73a76609bace9949ea8195d2ee8213cb3f into amd-gfx:300e9c581b42

Local branch amd-gfx 300e9c5 Manually merged main:bafda89a0944d947fc4b3b5663185e07a397ac30 into amd-gfx:2e4aa61a46dc
Remote branch main b590ba7 [NFC] Rename DependentModules  in ModuleFile to `TransitiveImports`

Author: SC llvm team

clang/include/clang/Serialization/ASTReader.h

@@ -601,11 +601,11 @@ class ASTReader
 
   /// An array of lexical contents of a declaration context, as a sequence of
   /// Decl::Kind, DeclID pairs.
-  using unalighed_decl_id_t =
+  using unaligned_decl_id_t =
       llvm::support::detail::packed_endian_specific_integral<
           serialization::DeclID, llvm::endianness::native,
           llvm::support::unaligned>;
-  using LexicalContents = ArrayRef<unalighed_decl_id_t>;
+  using LexicalContents = ArrayRef<unaligned_decl_id_t>;
 
   /// Map from a DeclContext to its lexical contents.
   llvm::DenseMap<const DeclContext*, std::pair<ModuleFile*, LexicalContents>>
@@ -2246,7 +2246,7 @@ class ASTReader
 
     auto [Loc, ModuleFileIndex] = ReadUntranslatedSourceLocation(Raw, Seq);
     ModuleFile *OwningModuleFile =
-        ModuleFileIndex == 0 ? &MF : MF.DependentModules[ModuleFileIndex - 1];
+        ModuleFileIndex == 0 ? &MF : MF.TransitiveImports[ModuleFileIndex - 1];
 
     assert(!SourceMgr.isLoadedSourceLocation(Loc) &&
            "Run out source location space");

clang/include/clang/Serialization/ModuleFile.h

@@ -513,11 +513,11 @@ class ModuleFile {
 
   /// List of modules which this modules dependent on. Different
   /// from `Imports`, this includes indirectly imported modules too.
-  /// The order of DependentModules is significant. It should keep
+  /// The order of TransitiveImports is significant. It should keep
   /// the same order with that module file manager when we write
   /// the current module file. The value of the member will be initialized
   /// in `ASTReader::ReadModuleOffsetMap`.
-  llvm::SmallVector<ModuleFile *, 16> DependentModules;
+  llvm::SmallVector<ModuleFile *, 16> TransitiveImports;
 
   /// Determine whether this module was directly imported at
   /// any point during translation.

clang/lib/Serialization/ASTReader.cpp

@@ -1264,7 +1264,7 @@ bool ASTReader::ReadLexicalDeclContextStorage(ModuleFile &M,
   if (!Lex.first) {
     Lex = std::make_pair(
         &M, llvm::ArrayRef(
-                reinterpret_cast<const unalighed_decl_id_t *>(Blob.data()),
+                reinterpret_cast<const unaligned_decl_id_t *>(Blob.data()),
                 Blob.size() / sizeof(DeclID)));
   }
   DC->setHasExternalLexicalStorage(true);
@@ -3401,7 +3401,7 @@ llvm::Error ASTReader::ReadASTBlock(ModuleFile &F,
     case TU_UPDATE_LEXICAL: {
       DeclContext *TU = ContextObj->getTranslationUnitDecl();
       LexicalContents Contents(
-          reinterpret_cast<const unalighed_decl_id_t *>(Blob.data()),
+          reinterpret_cast<const unaligned_decl_id_t *>(Blob.data()),
           static_cast<unsigned int>(Blob.size() / sizeof(DeclID)));
       TULexicalDecls.push_back(std::make_pair(&F, Contents));
       TU->setHasExternalLexicalStorage(true);
@@ -4059,7 +4059,7 @@ void ASTReader::ReadModuleOffsetMap(ModuleFile &F) const {
   RemapBuilder DeclRemap(F.DeclRemap);
   RemapBuilder TypeRemap(F.TypeRemap);
 
-  auto &ImportedModuleVector = F.DependentModules;
+  auto &ImportedModuleVector = F.TransitiveImports;
   assert(ImportedModuleVector.empty());
 
   while (Data < DataEnd) {

llvm/include/llvm/Config/llvm-config.h.cmake

@@ -16,7 +16,7 @@
 
 /* Indicate that this is LLVM compiled from the amd-gfx branch. */
 #define LLVM_HAVE_BRANCH_AMD_GFX
-#define LLVM_MAIN_REVISION 499824
+#define LLVM_MAIN_REVISION 499830
 
 /* Define if LLVM_ENABLE_DUMP is enabled */
 #cmakedefine LLVM_ENABLE_DUMP

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -1801,8 +1801,11 @@ void DAGCombiner::Run(CombineLevel AtLevel) {
 
     if (N->getNumValues() == RV->getNumValues())
       DAG.ReplaceAllUsesWith(N, RV.getNode());
-    else
+    else {
+      assert(N->getValueType(0) == RV.getValueType() &&
+             N->getNumValues() == 1 && "Type mismatch");
       DAG.ReplaceAllUsesWith(N, &RV);
+    }
 
     // Push the new node and any users onto the worklist.  Omit this if the
     // new node is the EntryToken (e.g. if a store managed to get optimized

llvm/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp

@@ -723,9 +723,10 @@ void X86AsmBackend::applyFixup(const MCAssembler &Asm, const MCFixup &Fixup,
 bool X86AsmBackend::mayNeedRelaxation(const MCInst &MI,
                                       const MCSubtargetInfo &STI) const {
   unsigned Opcode = MI.getOpcode();
+  unsigned SkipOperands = X86::isCCMPCC(Opcode) ? 2 : 0;
   return isRelaxableBranch(Opcode) ||
          (X86::getOpcodeForLongImmediateForm(Opcode) != Opcode &&
-          MI.getOperand(MI.getNumOperands() - 1).isExpr());
+          MI.getOperand(MI.getNumOperands() - 1 - SkipOperands).isExpr());
 }
 
 bool X86AsmBackend::fixupNeedsRelaxation(const MCFixup &Fixup,

llvm/lib/Target/X86/MCTargetDesc/X86EncodingOptimizationForImmediate.def

@@ -18,6 +18,12 @@ ENTRY(CMP32mi, CMP32mi8)
 ENTRY(CMP32ri, CMP32ri8)
 ENTRY(CMP64mi32, CMP64mi8)
 ENTRY(CMP64ri32, CMP64ri8)
+ENTRY(CCMP16mi, CCMP16mi8)
+ENTRY(CCMP16ri, CCMP16ri8)
+ENTRY(CCMP32mi, CCMP32mi8)
+ENTRY(CCMP32ri, CCMP32ri8)
+ENTRY(CCMP64mi32, CCMP64mi8)
+ENTRY(CCMP64ri32, CCMP64ri8)
 ENTRY(PUSH16i, PUSH16i8)
 ENTRY(PUSH32i, PUSH32i8)
 ENTRY(PUSH64i32, PUSH64i8)

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -49239,8 +49239,7 @@ static SDValue combineX86SubCmpForFlags(SDNode *N, SDValue Flag,
 
   SDValue SetCC = N->getOperand(0);
 
-  // TODO: Remove the check hasCCMP() and update the non-APX tests.
-  if (!ST.hasCCMP() || SetCC.getOpcode() != X86ISD::SETCC || !Flag.hasOneUse())
+  if (SetCC.getOpcode() != X86ISD::SETCC || !Flag.hasOneUse())
     return SDValue();
 
   // Check the only user of flag is `brcond ne`.
@@ -49253,8 +49252,10 @@ static SDValue combineX86SubCmpForFlags(SDNode *N, SDValue Flag,
     return SDValue();
 
   SDValue X = SetCC.getOperand(1);
-  // Replace API is called manually here b/c the number of results may change.
-  DAG.ReplaceAllUsesOfValueWith(Flag, X);
+  // sub has two results while X only have one. DAG combine assumes the value
+  // type matches.
+  if (N->getOpcode() == X86ISD::SUB)
+    X = DAG.getMergeValues({N->getOperand(0), X}, SDLoc(N));
 
   SDValue CCN = SetCC.getOperand(0);
   X86::CondCode CC =

llvm/lib/Target/X86/X86Subtarget.cpp

@@ -252,7 +252,7 @@ void X86Subtarget::initSubtargetFeatures(StringRef CPU, StringRef TuneCPU,
     CPU = "generic";
 
   if (TuneCPU.empty())
-    TuneCPU = HasX86_64 ? "generic" : "i586";
+    TuneCPU = "i586"; // FIXME: "generic" is more modern than llc tests expect.
 
   std::string FullFS = X86_MC::ParseX86Triple(TargetTriple);
   assert(!FullFS.empty() && "Failed to parse X86 triple");

llvm/test/MC/X86/apx/ccmp-reloc.s

@@ -0,0 +1,14 @@
+// RUN: llvm-mc -triple x86_64-linux-gnu -filetype=obj %s | llvm-readobj -r - | FileCheck %s
+
+// CHECK:      Relocations [
+// CHECK-NEXT:   Section ({{[0-9]+}}) .rela.text {
+ccmpbb	{dfv=of}	$foo, %bl               // CHECK-NEXT:     R_X86_64_8
+ccmpbb	{dfv=of}	$foo, 123(%r8,%rax,4)   // CHECK-NEXT:     R_X86_64_8
+ccmpbw	{dfv=of}	$foo, %bx               // CHECK-NEXT:     R_X86_64_16
+ccmpbw	{dfv=of}	$foo, 123(%r8,%rax,4)   // CHECK-NEXT:     R_X86_64_16
+ccmpbl	{dfv=of}	$foo, %ebx              // CHECK-NEXT:     R_X86_64_32
+ccmpbl	{dfv=of}	$foo, 123(%r8,%rax,4)   // CHECK-NEXT:     R_X86_64_32
+ccmpbq	{dfv=of}	$foo, %rbx              // CHECK-NEXT:     R_X86_64_32S
+ccmpbq	{dfv=of}	$foo, 123(%r8,%rax,4)   // CHECK-NEXT:     R_X86_64_32S
+// CHECK-NEXT:   }
+// CHECK-NEXT: ]