merge main into amd-main

.ci/compute_projects.py

@@ -98,6 +98,12 @@
     "openmp",  # https://github.com/google/llvm-premerge-checks/issues/410
 }
 
+# Runtimes configured for cross-compilation using LLVM_RUNTIME_TARGETS.
+# The same build may also use LLVM_ENABLE_RUNTIMES for other runtimes.
+CROSS_COMPILATION_RUNTIMES = {
+    "libclc",
+}
+
 EXCLUDE_WINDOWS = {
     "cross-project-tests",  # TODO(issues/132797): Tests are failing.
     "openmp",  # TODO(issues/132799): Does not detect perl installation.
@@ -147,7 +153,7 @@
     "flang": "check-flang",
     "flang-rt": "check-flang-rt",
     "libc": "check-libc",
-    "libclc": "check-libclc",
+    "libclc": "check-libclc-amdgcn-amd-amdhsa-llvm",
     "lld": "check-lld",
     "lldb": "check-lldb",
     "mlir": "check-mlir",
@@ -276,7 +282,9 @@ def _compute_runtimes_to_build(
     for modified_project in modified_projects:
         if modified_project in DEPENDENT_RUNTIMES_TO_BUILD:
             runtimes_to_build.update(DEPENDENT_RUNTIMES_TO_BUILD[modified_project])
-    return _exclude_projects(runtimes_to_build, platform)
+    runtimes_to_build = _exclude_projects(runtimes_to_build, platform)
+    runtimes_to_build -= CROSS_COMPILATION_RUNTIMES
+    return runtimes_to_build
 
 
 def _path_matches(matcher: tuple[str], file_path: tuple[str]) -> bool:
@@ -320,7 +328,10 @@ def get_env_variables(modified_files: list[str], platform: str) -> Set[str]:
     runtimes_to_build = _compute_runtimes_to_build(
         runtimes_to_test | runtimes_to_test_needs_reconfig, modified_projects, platform
     )
-    projects_to_build = _compute_projects_to_build(projects_to_test, runtimes_to_build)
+    cross_runtimes_to_test = runtimes_to_test & CROSS_COMPILATION_RUNTIMES
+    projects_to_build = _compute_projects_to_build(
+        projects_to_test, runtimes_to_build | cross_runtimes_to_test
+    )
     projects_check_targets = _compute_project_check_targets(projects_to_test)
     runtimes_check_targets = _compute_project_check_targets(runtimes_to_test)
     runtimes_check_targets_needs_reconfig = _compute_project_check_targets(

.ci/compute_projects_test.py

@@ -265,8 +265,11 @@ def test_include_libclc_in_runtimes(self):
         )
         self.assertEqual(env_variables["projects_to_build"], "clang;llvm")
         self.assertEqual(env_variables["project_check_targets"], "")
-        self.assertEqual(env_variables["runtimes_to_build"], "libclc")
-        self.assertEqual(env_variables["runtimes_check_targets"], "check-libclc")
+        self.assertEqual(env_variables["runtimes_to_build"], "")
+        self.assertEqual(
+            env_variables["runtimes_check_targets"],
+            "check-libclc-amdgcn-amd-amdhsa-llvm",
+        )
         self.assertEqual(env_variables["runtimes_check_targets_needs_reconfig"], "")
 
     def test_exclude_docs(self):
@@ -303,11 +306,11 @@ def test_ci(self):
         )
         self.assertEqual(
             env_variables["runtimes_to_build"],
-            "compiler-rt;flang-rt;libc;libclc;libcxx;libcxxabi;libunwind",
+            "compiler-rt;flang-rt;libc;libcxx;libcxxabi;libunwind",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets"],
-            "check-compiler-rt check-flang-rt check-libc check-libclc",
+            "check-compiler-rt check-flang-rt check-libc check-libclc-amdgcn-amd-amdhsa-llvm",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets_needs_reconfig"],
@@ -328,11 +331,11 @@ def test_windows_ci(self):
         )
         self.assertEqual(
             env_variables["runtimes_to_build"],
-            "compiler-rt;libclc",
+            "compiler-rt",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets"],
-            "check-compiler-rt check-libclc",
+            "check-compiler-rt check-libclc-amdgcn-amd-amdhsa-llvm",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets_needs_reconfig"],
@@ -377,11 +380,11 @@ def test_premerge_workflow(self):
         )
         self.assertEqual(
             env_variables["runtimes_to_build"],
-            "compiler-rt;flang-rt;libc;libclc;libcxx;libcxxabi;libunwind",
+            "compiler-rt;flang-rt;libc;libcxx;libcxxabi;libunwind",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets"],
-            "check-compiler-rt check-flang-rt check-libc check-libclc",
+            "check-compiler-rt check-flang-rt check-libc check-libclc-amdgcn-amd-amdhsa-llvm",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets_needs_reconfig"],
@@ -412,11 +415,11 @@ def test_third_party_benchmark(self):
         )
         self.assertEqual(
             env_variables["runtimes_to_build"],
-            "compiler-rt;flang-rt;libc;libclc;libcxx;libcxxabi;libunwind",
+            "compiler-rt;flang-rt;libc;libcxx;libcxxabi;libunwind",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets"],
-            "check-compiler-rt check-flang-rt check-libc check-libclc",
+            "check-compiler-rt check-flang-rt check-libc check-libclc-amdgcn-amd-amdhsa-llvm",
         )
         self.assertEqual(
             env_variables["runtimes_check_targets_needs_reconfig"],

.ci/monolithic-linux.sh

@@ -31,6 +31,14 @@ enable_cir="${6}"
 
 lit_args="-v --xunit-xml-output ${BUILD_DIR}/test-results.xml --use-unique-output-file-name --timeout=1200 --time-tests --succinct"
 
+runtime_cmake_args=()
+if [[ " ${runtime_targets} " == *" check-libclc-amdgcn-amd-amdhsa-llvm "* ]]; then
+  runtime_cmake_args+=(
+    -D RUNTIMES_amdgcn-amd-amdhsa-llvm_LLVM_ENABLE_RUNTIMES=libclc
+    -D LLVM_RUNTIME_TARGETS="default;amdgcn-amd-amdhsa-llvm"
+  )
+fi
+
 start-group "CMake"
 
 # Set the system llvm-symbolizer as preferred.
@@ -56,14 +64,14 @@ cmake -S "${MONOREPO_ROOT}"/llvm -B "${BUILD_DIR}" \
       -D CMAKE_CXX_COMPILER_LAUNCHER=sccache \
       -D CMAKE_DISABLE_PRECOMPILE_HEADERS=ON \
       -D LIBCXX_CXX_ABI=libcxxabi \
-      -D LIBCLC_TARGETS_TO_BUILD="amdgcn-amd-amdhsa-llvm" \
       -D MLIR_ENABLE_BINDINGS_PYTHON=ON \
       -D LLDB_ENABLE_PYTHON=ON \
       -D LLDB_ENFORCE_STRICT_TEST_REQUIREMENTS=ON \
       -D CMAKE_INSTALL_PREFIX="${INSTALL_DIR}" \
       -D CMAKE_EXE_LINKER_FLAGS="-no-pie" \
       -D LLVM_ENABLE_WERROR=ON \
-      -D LLVM_BINUTILS_INCDIR=/usr
+      -D LLVM_BINUTILS_INCDIR=/usr \
+      "${runtime_cmake_args[@]}"
 
 start-group "ninja"
 

.ci/monolithic-windows.sh

@@ -20,6 +20,14 @@ targets="${2}"
 runtimes="${3}"
 runtimes_targets="${4}"
 
+runtime_cmake_args=()
+if [[ " ${runtimes_targets} " == *" check-libclc-amdgcn-amd-amdhsa-llvm "* ]]; then
+  runtime_cmake_args+=(
+    -D RUNTIMES_amdgcn-amd-amdhsa-llvm_LLVM_ENABLE_RUNTIMES=libclc
+    -D LLVM_RUNTIME_TARGETS="default;amdgcn-amd-amdhsa-llvm"
+  )
+fi
+
 start-group "CMake"
 pip install -q -r "${MONOREPO_ROOT}"/.ci/all_requirements.txt
 
@@ -45,11 +53,11 @@ cmake -S "${MONOREPO_ROOT}"/llvm -B "${BUILD_DIR}" \
       -D CMAKE_CXX_COMPILER_LAUNCHER=sccache \
       -D CMAKE_DISABLE_PRECOMPILE_HEADERS=ON \
       -D MLIR_ENABLE_BINDINGS_PYTHON=ON \
-      -D LIBCLC_TARGETS_TO_BUILD="amdgcn-amd-amdhsa-llvm" \
       -D CMAKE_EXE_LINKER_FLAGS="/MANIFEST:NO" \
       -D CMAKE_MODULE_LINKER_FLAGS="/MANIFEST:NO" \
       -D CMAKE_SHARED_LINKER_FLAGS="/MANIFEST:NO" \
-      -D LLVM_ENABLE_RUNTIMES="${runtimes}"
+      -D LLVM_ENABLE_RUNTIMES="${runtimes}" \
+      "${runtime_cmake_args[@]}"
 
 start-group "ninja"
 

bolt/include/bolt/Passes/DataflowAnalysis.h

@@ -404,10 +404,12 @@ class DataflowAnalysis {
       // Propagate information from first instruction down to the last one
       StateTy *PrevState = &St;
       const MCInst *LAST = nullptr;
-      if (!Backward)
-        LAST = &*BB->rbegin();
-      else
-        LAST = &*BB->begin();
+      if (!BB->empty()) {
+        if (!Backward)
+          LAST = &*BB->rbegin();
+        else
+          LAST = &*BB->begin();
+      }
 
       auto doNext = [&](MCInst &Inst, const BinaryBasicBlock &BB) {
         StateTy CurState = derived().computeNext(Inst, *PrevState);

bolt/lib/Passes/AllocCombiner.cpp

@@ -64,6 +64,7 @@ static void runForAllWeCare(std::map<uint64_t, BinaryFunction> &BFs,
 void AllocCombinerPass::combineAdjustments(BinaryFunction &BF) {
   BinaryContext &BC = BF.getBinaryContext();
   for (BinaryBasicBlock &BB : BF) {
+    SmallVector<MCInst *, 2> ToErase;
     MCInst *Prev = nullptr;
     for (MCInst &Inst : llvm::reverse(BB)) {
       if (isIndifferentToSP(Inst, BC))
@@ -94,12 +95,14 @@ void AllocCombinerPass::combineAdjustments(BinaryFunction &BF) {
         Inst.dump();
       });
 
-      BB.eraseInstruction(BB.findInstruction(Prev));
+      ToErase.push_back(Prev);
       ++NumCombined;
       DynamicCountCombined += BB.getKnownExecutionCount();
       FuncsChanged.insert(&BF);
       Prev = &Inst;
     }
+    for (MCInst *Inst : ToErase)
+      BB.eraseInstruction(BB.findInstruction(Inst));
   }
 }
 

bolt/lib/Passes/IndirectCallPromotion.cpp

@@ -775,8 +775,7 @@ IndirectCallPromotion::rewriteCall(
   InstructionListType TailInsts;
   const MCInst *TailInst = &CallInst;
   if (IsTailCallOrJT)
-    while (TailInst + 1 < &(*IndCallBlock.end()) &&
-           MIB->isPseudo(*(TailInst + 1)))
+    while (TailInst != &IndCallBlock.back() && MIB->isPseudo(*(TailInst + 1)))
       TailInsts.push_back(*++TailInst);
 
   InstructionListType MovedInst = IndCallBlock.splitInstructions(&CallInst);

bolt/lib/Passes/ShrinkWrapping.cpp

@@ -1887,13 +1887,16 @@ Expected<bool> ShrinkWrapping::processInsertions() {
 void ShrinkWrapping::processDeletions() {
   LivenessAnalysis &LA = Info.getLivenessAnalysis();
   for (BinaryBasicBlock &BB : BF) {
-    for (auto II = BB.begin(); II != BB.end(); ++II) {
+    for (auto II = BB.begin(); II != BB.end();) {
       MCInst &Inst = *II;
       auto TodoList = BC.MIB->tryGetAnnotationAs<std::vector<WorklistItem>>(
           Inst, getAnnotationIndex());
-      if (!TodoList)
+      if (!TodoList) {
+        ++II;
         continue;
+      }
       // Process all deletions
+      bool Erased = false;
       for (WorklistItem &Item : *TodoList) {
         if (Item.Action != WorklistItem::Erase &&
             Item.Action != WorklistItem::ChangeToAdjustment)
@@ -1916,9 +1919,12 @@ void ShrinkWrapping::processDeletions() {
           dbgs() << "Erasing: ";
           BC.printInstruction(dbgs(), Inst);
         });
-        II = std::prev(BB.eraseInstruction(II));
+        II = BB.eraseInstruction(II);
+        Erased = true;
         break;
       }
+      if (!Erased)
+        ++II;
     }
   }
 }

bolt/lib/Passes/TailDuplication.cpp

@@ -167,18 +167,22 @@ void TailDuplication::constantAndCopyPropagate(
 
   BlocksToPropagate.insert(BlocksToPropagate.begin(), &OriginalBB);
   // Iterate through the original instructions to find one to propagate
-  for (auto Itr = OriginalBB.begin(); Itr != OriginalBB.end(); ++Itr) {
+  for (auto Itr = OriginalBB.begin(); Itr != OriginalBB.end();) {
     MCInst &OriginalInst = *Itr;
     // It must be a non conditional
-    if (BC.MIB->isConditionalMove(OriginalInst))
+    if (BC.MIB->isConditionalMove(OriginalInst)) {
+      ++Itr;
       continue;
+    }
 
     // Move immediate or move register
     if ((!BC.MII->get(OriginalInst.getOpcode()).isMoveImmediate() ||
          !OriginalInst.getOperand(1).isImm()) &&
         (!BC.MII->get(OriginalInst.getOpcode()).isMoveReg() ||
-         !OriginalInst.getOperand(1).isReg()))
+         !OriginalInst.getOperand(1).isReg())) {
+      ++Itr;
       continue;
+    }
 
     // True if this is constant propagation and not copy propagation
     bool ConstantProp = BC.MII->get(OriginalInst.getOpcode()).isMoveImmediate();
@@ -247,7 +251,9 @@ void TailDuplication::constantAndCopyPropagate(
       // to replace is active for constant propagation
       StaticInstructionDeletionCount++;
       DynamicInstructionDeletionCount += OriginalBB.getExecutionCount();
-      Itr = std::prev(OriginalBB.eraseInstruction(Itr));
+      Itr = OriginalBB.eraseInstruction(Itr);
+    } else {
+      ++Itr;
     }
   }
 }