Merge branch 'main' into review/benchmarks-refactor-sequence-containers

Author: ldionne

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

@@ -108,7 +108,7 @@ RUN choco install -y handle
 
 RUN pip3 install pywin32 buildbot-worker==2.8.4
 
-ARG RUNNER_VERSION=2.319.1
+ARG RUNNER_VERSION=2.321.0
 ENV RUNNER_VERSION=$RUNNER_VERSION
 
 RUN powershell -Command \

.github/workflows/premerge.yaml

@@ -30,6 +30,8 @@ jobs:
           fetch-depth: 2
       - name: Setup ccache
         uses: hendrikmuhs/[email protected]
+        with:
+          max-size: "2000M"
       - name: Build and Test
         # Mark the job as a success even if the step fails so that people do
         # not get notified while the new premerge pipeline is in an

bolt/CMakeLists.txt

@@ -163,8 +163,8 @@ if (BOLT_ENABLE_RUNTIME)
   add_llvm_install_targets(install-bolt_rt
     DEPENDS bolt_rt bolt
     COMPONENT bolt)
-  set(LIBBOLT_RT_INSTR "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib/libbolt_rt_instr.a")
-  set(LIBBOLT_RT_HUGIFY "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib/libbolt_rt_hugify.a")
+  set(LIBBOLT_RT_INSTR "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_instr.a")
+  set(LIBBOLT_RT_HUGIFY "${CMAKE_CURRENT_BINARY_DIR}/bolt_rt-bins/lib${LLVM_LIBDIR_SUFFIX}/libbolt_rt_hugify.a")
 endif()
 
 find_program(GNU_LD_EXECUTABLE NAMES ${LLVM_DEFAULT_TARGET_TRIPLE}-ld.bfd ld.bfd DOC "GNU ld")

bolt/include/bolt/Core/BinaryContext.h

@@ -1363,6 +1363,12 @@ class BinaryContext {
     if (std::optional<uint32_t> Size = MIB->getSize(Inst))
       return *Size;
 
+    if (MIB->isPseudo(Inst))
+      return 0;
+
+    if (std::optional<uint32_t> Size = MIB->getInstructionSize(Inst))
+      return *Size;
+
     if (!Emitter)
       Emitter = this->MCE.get();
     SmallString<256> Code;

bolt/include/bolt/Core/MCPlusBuilder.h

@@ -1204,6 +1204,11 @@ class MCPlusBuilder {
   /// Get instruction size specified via annotation.
   std::optional<uint32_t> getSize(const MCInst &Inst) const;
 
+  /// Get target-specific instruction size.
+  virtual std::optional<uint32_t> getInstructionSize(const MCInst &Inst) const {
+    return std::nullopt;
+  }
+
   /// Set instruction size.
   void setSize(MCInst &Inst, uint32_t Size) const;
 

bolt/lib/Passes/Inliner.cpp

@@ -310,13 +310,13 @@ Inliner::inlineCall(BinaryBasicBlock &CallerBB,
       if (MIB.isPseudo(Inst))
         continue;
 
-      MIB.stripAnnotations(Inst, /*KeepTC=*/BC.isX86());
+      MIB.stripAnnotations(Inst, /*KeepTC=*/BC.isX86() || BC.isAArch64());
 
       // Fix branch target. Strictly speaking, we don't have to do this as
       // targets of direct branches will be fixed later and don't matter
       // in the CFG state. However, disassembly may look misleading, and
       // hence we do the fixing.
-      if (MIB.isBranch(Inst)) {
+      if (MIB.isBranch(Inst) && !MIB.isTailCall(Inst)) {
         assert(!MIB.isIndirectBranch(Inst) &&
                "unexpected indirect branch in callee");
         const BinaryBasicBlock *TargetBB =