Merged main:344228ebf45f9bd1f7626fdcd3c0fada0f0c8385 into amd-gfx:d6bc6d5dbcb2

Local branch amd-gfx d6bc6d5 Merge main into amd-gfx
Remote branch main 344228e [BOLT] Drop macro-fusion alignment (llvm#97358)

Author: SC llvm team

.github/CODEOWNERS

@@ -120,6 +120,9 @@ clang/test/AST/Interp/ @tbaederr
 /mlir/**/LLVMIR/**/BasicPtxBuilderInterface* @grypp
 /mlir/**/NVVM* @grypp
 
+# MLIR Index Dialect
+/mlir/**/Index* @mogball
+
 # MLIR Python Bindings
 /mlir/test/python/ @ftynse @makslevental @stellaraccident
 /mlir/python/ @ftynse @makslevental @stellaraccident

.github/workflows/issue-write.yml

@@ -5,6 +5,7 @@ on:
     workflows:
       - "Check code formatting"
       - "Check for private emails used in PRs"
+      - "PR Request Release Note"
     types:
       - completed
 
@@ -17,7 +18,11 @@ jobs:
     permissions:
       pull-requests: write
     if: >
-      github.event.workflow_run.event == 'pull_request'
+      github.event.workflow_run.event == 'pull_request' &&
+      (
+        github.event.workflow_run.conclusion == 'success' ||
+        github.event.workflow_run.conclusion == 'failure'
+      )
     steps:
       - name: 'Download artifact'
         uses: actions/download-artifact@6b208ae046db98c579e8a3aa621ab581ff575935 # v4.1.1
@@ -92,7 +97,11 @@ jobs:
 
             var pr_number = 0;
             gql_result.repository.ref.associatedPullRequests.nodes.forEach((pr) => {
-              if (pr.baseRepository.owner.login = context.repo.owner && pr.state == 'OPEN') {
+
+              // The largest PR number is the one we care about.  The only way
+              // to have more than one associated pull requests is if all the
+              // old pull requests are in the closed state.
+              if (pr.baseRepository.owner.login = context.repo.owner && pr.number > pr_number) {
                 pr_number = pr.number;
               }
             });

.github/workflows/pr-request-release-note.yml

@@ -2,7 +2,6 @@ name: PR Request Release Note
 
 permissions:
   contents: read
-  pull-requests: write
 
 on:
   pull_request:
@@ -41,3 +40,10 @@ jobs:
             --token "$GITHUB_TOKEN" \
             request-release-note \
             --pr-number ${{ github.event.pull_request.number}}
+
+      - uses: actions/upload-artifact@26f96dfa697d77e81fd5907df203aa23a56210a8 #v4.3.0
+        if: always()
+        with:
+          name: workflow-args
+          path: |
+            comments

bolt/docs/CommandLineArgumentReference.md

@@ -259,6 +259,10 @@
 
   Always use long jumps/nops for Linux kernel static keys
 
+- `--match-profile-with-function-hash`
+
+  Match profile with function hash
+
 - `--max-data-relocations=<uint>`
 
   Maximum number of data relocations to process

bolt/include/bolt/Core/BinaryBasicBlock.h

@@ -842,15 +842,6 @@ class BinaryBasicBlock {
   bool analyzeBranch(const MCSymbol *&TBB, const MCSymbol *&FBB,
                      MCInst *&CondBranch, MCInst *&UncondBranch);
 
-  /// Return true if iterator \p I is pointing to the first instruction in
-  /// a pair that could be macro-fused.
-  bool isMacroOpFusionPair(const_iterator I) const;
-
-  /// If the basic block has a pair of instructions suitable for macro-fusion,
-  /// return iterator to the first instruction of the pair.
-  /// Otherwise return end().
-  const_iterator getMacroOpFusionPair() const;
-
   /// Printer required for printing dominator trees.
   void printAsOperand(raw_ostream &OS, bool PrintType = true) {
     if (PrintType)

bolt/include/bolt/Core/BinaryContext.h

@@ -698,10 +698,6 @@ class BinaryContext {
 
   /// Binary-wide aggregated stats.
   struct BinaryStats {
-    /// Stats for macro-fusion.
-    uint64_t MissedMacroFusionPairs{0};
-    uint64_t MissedMacroFusionExecCount{0};
-
     /// Stats for stale profile matching:
     ///   the total number of basic blocks in the profile
     uint32_t NumStaleBlocks{0};

bolt/include/bolt/Core/BinaryFunction.h

@@ -835,10 +835,6 @@ class BinaryFunction {
   /// them.
   void calculateLoopInfo();
 
-  /// Calculate missed macro-fusion opportunities and update BinaryContext
-  /// stats.
-  void calculateMacroOpFusionStats();
-
   /// Returns if BinaryDominatorTree has been constructed for this function.
   bool hasDomTree() const { return BDT != nullptr; }
 
@@ -930,6 +926,10 @@ class BinaryFunction {
     return const_cast<BinaryFunction *>(this)->getInstructionAtOffset(Offset);
   }
 
+  /// When the function is in disassembled state, return an instruction that
+  /// contains the \p Offset.
+  MCInst *getInstructionContainingOffset(uint64_t Offset);
+
   std::optional<MCInst> disassembleInstructionAtOffset(uint64_t Offset) const;
 
   /// Return offset for the first instruction. If there is data at the

bolt/include/bolt/Passes/BinaryFunctionCallGraph.h renamed to bolt/include/bolt/Core/BinaryFunctionCallGraph.h

@@ -1,4 +1,4 @@
-//===- bolt/Passes/CallGraph.h ----------------------------------*- C++ -*-===//
+//===- bolt/Core/CallGraph.h ----------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -9,7 +9,7 @@
 #ifndef BOLT_PASSES_BINARY_FUNCTION_CALLGRAPH_H
 #define BOLT_PASSES_BINARY_FUNCTION_CALLGRAPH_H
 
-#include "bolt/Passes/CallGraph.h"
+#include "bolt/Core/CallGraph.h"
 #include <deque>
 #include <functional>
 #include <unordered_map>

bolt/include/bolt/Passes/CallGraph.h renamed to bolt/include/bolt/Core/CallGraph.h

@@ -1,4 +1,4 @@
-//===- bolt/Passes/CallGraph.h ----------------------------------*- C++ -*-===//
+//===- bolt/Core/CallGraph.h ----------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.

bolt/include/bolt/Passes/CallGraphWalker.h renamed to bolt/include/bolt/Core/CallGraphWalker.h

@@ -1,4 +1,4 @@
-//===- bolt/Passes/CallGraphWalker.h ----------------------------*- C++ -*-===//
+//===- bolt/Core/CallGraphWalker.h ----------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.

bolt/include/bolt/Core/MCPlusBuilder.h

@@ -439,7 +439,20 @@ class MCPlusBuilder {
   }
 
   /// Check whether this conditional branch can be reversed
-  virtual bool isReversibleBranch(const MCInst &Inst) const { return true; }
+  virtual bool isReversibleBranch(const MCInst &Inst) const {
+    assert(!isUnsupportedInstruction(Inst) && isConditionalBranch(Inst) &&
+           "Instruction is not known conditional branch");
+
+    if (isDynamicBranch(Inst))
+      return false;
+    return true;
+  }
+
+  /// Return true if this instruction inhibits analysis of the containing
+  /// function.
+  virtual bool isUnsupportedInstruction(const MCInst &Inst) const {
+    return false;
+  }
 
   /// Return true of the instruction is of pseudo kind.
   virtual bool isPseudo(const MCInst &Inst) const {
@@ -917,13 +930,6 @@ class MCPlusBuilder {
   /// Return true if the instruction is encoded using EVEX (AVX-512).
   virtual bool hasEVEXEncoding(const MCInst &Inst) const { return false; }
 
-  /// Return true if a pair of instructions represented by \p Insts
-  /// could be fused into a single uop.
-  virtual bool isMacroOpFusionPair(ArrayRef<MCInst> Insts) const {
-    llvm_unreachable("not implemented");
-    return false;
-  }
-
   struct X86MemOperand {
     unsigned BaseRegNum;
     int64_t ScaleImm;

bolt/include/bolt/Passes/HFSort.h

@@ -19,7 +19,7 @@
 #ifndef BOLT_PASSES_HFSORT_H
 #define BOLT_PASSES_HFSORT_H
 
-#include "bolt/Passes/CallGraph.h"
+#include "bolt/Core/CallGraph.h"
 
 #include <string>
 #include <vector>

bolt/include/bolt/Passes/RegReAssign.h

@@ -9,7 +9,7 @@
 #ifndef BOLT_PASSES_REGREASSIGN_H
 #define BOLT_PASSES_REGREASSIGN_H
 
-#include "bolt/Passes/BinaryFunctionCallGraph.h"
+#include "bolt/Core/BinaryFunctionCallGraph.h"
 #include "bolt/Passes/BinaryPasses.h"
 #include "bolt/Passes/RegAnalysis.h"
 

bolt/include/bolt/Passes/ReorderFunctions.h

@@ -9,7 +9,7 @@
 #ifndef BOLT_PASSES_REORDER_FUNCTIONS_H
 #define BOLT_PASSES_REORDER_FUNCTIONS_H
 
-#include "bolt/Passes/BinaryFunctionCallGraph.h"
+#include "bolt/Core/BinaryFunctionCallGraph.h"
 #include "bolt/Passes/BinaryPasses.h"
 
 namespace llvm {

bolt/include/bolt/Rewrite/DWARFRewriter.h

@@ -16,7 +16,6 @@
 #include "llvm/ADT/StringRef.h"
 #include "llvm/CodeGen/DIE.h"
 #include "llvm/DWP/DWP.h"
-#include "llvm/MC/MCAsmLayout.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/Support/ToolOutputFile.h"
 #include <cstdint>
@@ -183,7 +182,7 @@ class DWARFRewriter {
   void updateDebugInfo();
 
   /// Update stmt_list for CUs based on the new .debug_line \p Layout.
-  void updateLineTableOffsets(const MCAsmLayout &Layout);
+  void updateLineTableOffsets(const MCAssembler &Asm);
 
   uint64_t getDwoRangesBase(uint64_t DWOId) { return DwoRangesBase[DWOId]; }
 

bolt/lib/Core/BinaryBasicBlock.cpp

@@ -404,45 +404,6 @@ bool BinaryBasicBlock::analyzeBranch(const MCSymbol *&TBB, const MCSymbol *&FBB,
                             CondBranch, UncondBranch);
 }
 
-bool BinaryBasicBlock::isMacroOpFusionPair(const_iterator I) const {
-  auto &MIB = Function->getBinaryContext().MIB;
-  ArrayRef<MCInst> Insts = Instructions;
-  return MIB->isMacroOpFusionPair(Insts.slice(I - begin()));
-}
-
-BinaryBasicBlock::const_iterator
-BinaryBasicBlock::getMacroOpFusionPair() const {
-  if (!Function->getBinaryContext().isX86())
-    return end();
-
-  if (getNumNonPseudos() < 2 || succ_size() != 2)
-    return end();
-
-  auto RI = getLastNonPseudo();
-  assert(RI != rend() && "cannot have an empty block with 2 successors");
-
-  BinaryContext &BC = Function->getBinaryContext();
-
-  // Skip instruction if it's an unconditional branch following
-  // a conditional one.
-  if (BC.MIB->isUnconditionalBranch(*RI))
-    ++RI;
-
-  if (!BC.MIB->isConditionalBranch(*RI))
-    return end();
-
-  // Start checking with instruction preceding the conditional branch.
-  ++RI;
-  if (RI == rend())
-    return end();
-
-  auto II = std::prev(RI.base()); // convert to a forward iterator
-  if (isMacroOpFusionPair(II))
-    return II;
-
-  return end();
-}
-
 MCInst *BinaryBasicBlock::getTerminatorBefore(MCInst *Pos) {
   BinaryContext &BC = Function->getBinaryContext();
   auto Itr = rbegin();

bolt/lib/Core/BinaryContext.cpp

@@ -2422,14 +2422,14 @@ BinaryContext::calculateEmittedSize(BinaryFunction &BF, bool FixBranches) {
   // Obtain fragment sizes.
   std::vector<uint64_t> FragmentSizes;
   // Main fragment size.
-  const uint64_t HotSize =
-      Layout.getSymbolOffset(*EndLabel) - Layout.getSymbolOffset(*StartLabel);
+  const uint64_t HotSize = Assembler.getSymbolOffset(*EndLabel) -
+                           Assembler.getSymbolOffset(*StartLabel);
   FragmentSizes.push_back(HotSize);
   // Split fragment sizes.
   uint64_t ColdSize = 0;
   for (const auto &Labels : SplitLabels) {
-    uint64_t Size = Layout.getSymbolOffset(*Labels.second) -
-                    Layout.getSymbolOffset(*Labels.first);
+    uint64_t Size = Assembler.getSymbolOffset(*Labels.second) -
+                    Assembler.getSymbolOffset(*Labels.first);
     FragmentSizes.push_back(Size);
     ColdSize += Size;
   }
@@ -2439,7 +2439,8 @@ BinaryContext::calculateEmittedSize(BinaryFunction &BF, bool FixBranches) {
   for (FunctionFragment &FF : BF.getLayout().fragments()) {
     BinaryBasicBlock *PrevBB = nullptr;
     for (BinaryBasicBlock *BB : FF) {
-      const uint64_t BBStartOffset = Layout.getSymbolOffset(*(BB->getLabel()));
+      const uint64_t BBStartOffset =
+          Assembler.getSymbolOffset(*(BB->getLabel()));
       BB->setOutputStartAddress(BBStartOffset);
       if (PrevBB)
         PrevBB->setOutputEndAddress(BBStartOffset);

bolt/lib/Core/BinaryEmitter.cpp

@@ -38,19 +38,6 @@ extern cl::opt<bool> PreserveBlocksAlignment;
 cl::opt<bool> AlignBlocks("align-blocks", cl::desc("align basic blocks"),
                           cl::cat(BoltOptCategory));
 
-cl::opt<MacroFusionType>
-AlignMacroOpFusion("align-macro-fusion",
-  cl::desc("fix instruction alignment for macro-fusion (x86 relocation mode)"),
-  cl::init(MFT_HOT),
-  cl::values(clEnumValN(MFT_NONE, "none",
-               "do not insert alignment no-ops for macro-fusion"),
-             clEnumValN(MFT_HOT, "hot",
-               "only insert alignment no-ops on hot execution paths (default)"),
-             clEnumValN(MFT_ALL, "all",
-               "always align instructions to allow macro-fusion")),
-  cl::ZeroOrMore,
-  cl::cat(BoltRelocCategory));
-
 static cl::list<std::string>
 BreakFunctionNames("break-funcs",
   cl::CommaSeparated,
@@ -453,20 +440,7 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         Streamer.emitLabel(EntrySymbol);
     }
 
-    // Check if special alignment for macro-fusion is needed.
-    bool MayNeedMacroFusionAlignment =
-        (opts::AlignMacroOpFusion == MFT_ALL) ||
-        (opts::AlignMacroOpFusion == MFT_HOT && BB->getKnownExecutionCount());
-    BinaryBasicBlock::const_iterator MacroFusionPair;
-    if (MayNeedMacroFusionAlignment) {
-      MacroFusionPair = BB->getMacroOpFusionPair();
-      if (MacroFusionPair == BB->end())
-        MayNeedMacroFusionAlignment = false;
-    }
-
     SMLoc LastLocSeen;
-    // Remember if the last instruction emitted was a prefix.
-    bool LastIsPrefix = false;
     for (auto I = BB->begin(), E = BB->end(); I != E; ++I) {
       MCInst &Instr = *I;
 
@@ -479,16 +453,6 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         continue;
       }
 
-      // Handle macro-fusion alignment. If we emitted a prefix as
-      // the last instruction, we should've already emitted the associated
-      // alignment hint, so don't emit it twice.
-      if (MayNeedMacroFusionAlignment && !LastIsPrefix &&
-          I == MacroFusionPair) {
-        // This assumes the second instruction in the macro-op pair will get
-        // assigned to its own MCRelaxableFragment. Since all JCC instructions
-        // are relaxable, we should be safe.
-      }
-
       if (!EmitCodeOnly) {
         // A symbol to be emitted before the instruction to mark its location.
         MCSymbol *InstrLabel = BC.MIB->getInstLabel(Instr);
@@ -525,7 +489,6 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
       }
 
       Streamer.emitInstruction(Instr, *BC.STI);
-      LastIsPrefix = BC.MIB->isPrefix(Instr);
     }
   }