Merge branch 'llvm:main' into main

Author: tdupes

.arcconfig

@@ -869,6 +869,9 @@ backend:SystemZ:
 third-party:unittests:
   - third-party/unittests/**
 
+third-party:benchmark:
+  - third-party/benchmark/**
+
 llvm:binary-utilities:
   - llvm/docs/CommandGuide/llvm-*
   - llvm/include/llvm/BinaryFormat/**

.arclint

@@ -77,14 +77,18 @@ jobs:
           cp ./.github/workflows/containers/github-action-ci/storage.conf ~/.config/containers/storage.conf
           podman info
 
+        # Download the container image into /mnt/podman rather than
+        # $GITHUB_WORKSPACE to avoid space limitations on the default drive
+        # and use the permissions setup for /mnt/podman.
       - name: Download stage1-toolchain
         uses: actions/download-artifact@v4
         with:
           name: stage1-toolchain
+          path: /mnt/podman
 
       - name: Load stage1-toolchain
         run: |
-          podman load -i stage1-toolchain.tar
+          podman load -i /mnt/podman/stage1-toolchain.tar
 
       - name: Build Container
         working-directory: ./.github/workflows/containers/github-action-ci/

.github/new-prs-labeler.yml

@@ -12,11 +12,13 @@ COPY --from=stage2-toolchain $LLVM_SYSROOT $LLVM_SYSROOT
 # Need to install curl for hendrikmuhs/ccache-action
 # Need nodejs for some of the GitHub actions.
 # Need perl-modules for clang analyzer tests.
+# Need git for SPIRV-Tools tests.
 RUN apt-get update && \
     apt-get install -y \
     binutils \
     cmake \
     curl \
+    git \
     libstdc++-11-dev \
     ninja-build \
     nodejs \

.github/workflows/build-ci-container.yml

@@ -2056,6 +2056,14 @@ class BinaryFunction {
   /// Returns false if disassembly failed.
   Error disassemble();
 
+  /// An external interface to register a branch while the function is in
+  /// disassembled state. Allows to make custom modifications to the
+  /// disassembler. E.g., a pre-CFG pass can add an instruction and register
+  /// a branch that will later be used during the CFG construction.
+  ///
+  /// Return a label at the branch destination.
+  MCSymbol *registerBranch(uint64_t Src, uint64_t Dst);
+
   Error handlePCRelOperand(MCInst &Instruction, uint64_t Address,
                            uint64_t Size);
 

.github/workflows/containers/github-action-ci/stage2.Dockerfile

@@ -18,6 +18,7 @@
 
 #include "bolt/Core/MCPlusBuilder.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ThreadPool.h"
 
 using namespace llvm;
 
@@ -28,8 +29,6 @@ extern cl::opt<unsigned> TaskCount;
 } // namespace opts
 
 namespace llvm {
-class ThreadPool;
-
 namespace bolt {
 class BinaryContext;
 class BinaryFunction;

bolt/include/bolt/Core/BinaryFunction.h

@@ -1445,6 +1445,16 @@ Error BinaryFunction::disassemble() {
   return Error::success();
 }
 
+MCSymbol *BinaryFunction::registerBranch(uint64_t Src, uint64_t Dst) {
+  assert(CurrentState == State::Disassembled &&
+         "Cannot register branch unless function is in disassembled state.");
+  assert(containsAddress(Src) && containsAddress(Dst) &&
+         "Cannot register external branch.");
+  MCSymbol *Target = getOrCreateLocalLabel(Dst);
+  TakenBranches.emplace_back(Src - getAddress(), Dst - getAddress());
+  return Target;
+}
+
 bool BinaryFunction::scanExternalRefs() {
   bool Success = true;
   bool DisassemblyFailed = false;
@@ -1759,13 +1769,6 @@ void BinaryFunction::postProcessJumpTables() {
       }
     }
   }
-
-  // Remove duplicates branches. We can get a bunch of them from jump tables.
-  // Without doing jump table value profiling we don't have use for extra
-  // (duplicate) branches.
-  llvm::sort(TakenBranches);
-  auto NewEnd = std::unique(TakenBranches.begin(), TakenBranches.end());
-  TakenBranches.erase(NewEnd, TakenBranches.end());
 }
 
 bool BinaryFunction::validateExternallyReferencedOffsets() {
@@ -2128,6 +2131,13 @@ Error BinaryFunction::buildCFG(MCPlusBuilder::AllocatorIdTy AllocatorId) {
   // e.g. exit(3), etc. Otherwise we'll see a false fall-through
   // blocks.
 
+  // Remove duplicates branches. We can get a bunch of them from jump tables.
+  // Without doing jump table value profiling we don't have a use for extra
+  // (duplicate) branches.
+  llvm::sort(TakenBranches);
+  auto NewEnd = std::unique(TakenBranches.begin(), TakenBranches.end());
+  TakenBranches.erase(NewEnd, TakenBranches.end());
+
   for (std::pair<uint32_t, uint32_t> &Branch : TakenBranches) {
     LLVM_DEBUG(dbgs() << "registering branch [0x"
                       << Twine::utohexstr(Branch.first) << "] -> [0x"

bolt/include/bolt/Core/ParallelUtilities.h

@@ -381,6 +381,14 @@ static uint64_t encodeValueAArch64(uint64_t Type, uint64_t Value, uint64_t PC) {
     // OP 1001_01 goes in bits 31:26 of BL.
     Value = ((Value >> 2) & 0x3ffffff) | 0x94000000ULL;
     break;
+  case ELF::R_AARCH64_JUMP26:
+    Value -= PC;
+    assert(isInt<28>(Value) &&
+           "only PC +/- 128MB is allowed for direct branch");
+    // Immediate goes in bits 25:0 of B.
+    // OP 0001_01 goes in bits 31:26 of B.
+    Value = ((Value >> 2) & 0x3ffffff) | 0x14000000ULL;
+    break;
   }
   return Value;
 }