Merge branch 'main' of github.com:EnzymeAD/Enzyme into jmp/aa-benchma…

…rking

.devcontainer/devcontainer.json

@@ -1,8 +1,8 @@
 // available ubuntu versions: [20, 22]
-// available llvm versions: [9, 10, 11, 12, 13, 14, 15]
+// available llvm versions: [11, 12, 13, 14, 15, 16, 17, 18]
 {
  "name": "Enzyme",
- "image": "ghcr.io/enzymead/enzyme-dev-docker/ubuntu-20-llvm-12:latest",
+ "image": "ghcr.io/enzymead/enzyme-dev-docker/ubuntu-22-llvm-16:latest",
  "mounts": [
  "source=enzyme-bashhistory,target=/commandhistory,type=volume",
  "source=enzyme-extensions,target=/home/vscode/.vscode-server/extensions,type=volume",
@@ -14,7 +14,9 @@
  "customizations": {
  "vscode": {
  "extensions": [
- "ms-vscode.cpptools-extension-pack"
+ "llvm-vs-code-extensions.vscode-clangd",
+ "BazelBuild.vscode-bazel",
+ "twxs.cmake"
  ]
  }
  }

.github/dependabot.yml

@@ -0,0 +1,12 @@
+# To get started with Dependabot version updates, you'll need to specify which
+# package ecosystems to update and where the package manifests are located.
+# Please see the documentation for all configuration options:
+# https://docs.github.com/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file
+
+version: 2
+updates:
+# Maintain dependencies for GitHub Actions
+ - package-ecosystem: "github-actions"
+ directory: "/"
+ schedule:
+ interval: "monthly"

.github/workflows/bcload.yml

@@ -27,7 +27,7 @@ jobs:
  sudo sed -i 's/add_executable(llvm-omp-device-info IMPORTED)//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  sudo sed -i 's/llvm-omp-device-info//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  fi
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: mkdir
  run: cd enzyme && rm -rf build && mkdir build
  - name: cmake

.github/workflows/benchmark.yml

@@ -37,7 +37,7 @@ jobs:
  sudo sed -i 's/add_executable(llvm-omp-device-info IMPORTED)//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  sudo sed -i 's/llvm-omp-device-info//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  fi
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: mkdir
  run: rm -rf build && mkdir build
  - name: cmake

.github/workflows/ccpp.yml

@@ -27,14 +27,15 @@ jobs:
  - name: add llvm 
  run: |
  wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key|sudo apt-key add -
+ sudo apt-get install -y libmpfr-dev
  sudo apt-add-repository "deb http://apt.llvm.org/`lsb_release -c | cut -f2`/ llvm-toolchain-`lsb_release -c | cut -f2`-${{ matrix.llvm }} main" || true
  sudo apt-get install -y cmake gcc g++ llvm-${{ matrix.llvm }}-dev libomp-${{ matrix.llvm }}-dev lld-${{ matrix.llvm }} clang-${{ matrix.llvm }} libclang-${{ matrix.llvm }}-dev libeigen3-dev libboost-dev libzstd-dev
  sudo python3 -m pip install --upgrade pip lit
  if [[ '${{ matrix.llvm }}' == '13' ]]; then
  sudo sed -i 's/add_executable(llvm-omp-device-info IMPORTED)//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  sudo sed -i 's/llvm-omp-device-info//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  fi
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: mkdir
  run: rm -rf build && mkdir build
  - name: cmake

.github/workflows/doxygen.yml

@@ -9,9 +9,9 @@ jobs:
  docs:
  runs-on: ubuntu-20.04
  steps:
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
 
- - uses: mattnotmitt/doxygen-action@v1.9.2
+ - uses: mattnotmitt/doxygen-action@v1.9.8
  with:
  working-directory: 'enzyme/'
  doxyfile-path: 'doxygen.cfg'

.github/workflows/enzyme-ci.yml

@@ -32,7 +32,7 @@ jobs:
  sudo sed -i 's/add_executable(llvm-omp-device-info IMPORTED)//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  sudo sed -i 's/llvm-omp-device-info//g' /usr/lib/llvm-${{matrix.llvm}}/lib/cmake/llvm/LLVMExports*.cmake
  fi
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: mkdir
  run: rm -rf build && mkdir build
  - name: cmake
@@ -61,7 +61,7 @@ jobs:
  strategy:
  fail-fast: false
  matrix:
- llvm: ["11", "12", "13", "14", "15"]
+ llvm: ["12", "13", "14", "15", "16"]
  build: ["Release", "Debug"] # "RelWithDebInfo"
 
  timeout-minutes: 30 
@@ -71,7 +71,7 @@ jobs:
  brew update
  brew install llvm@${{ matrix.llvm }} make cmake
  sudo python3 -m pip install --upgrade pip lit requests
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: mkdir
  run: rm -rf build && mkdir build
  - name: cmake
@@ -109,7 +109,7 @@ jobs:
  run: |
  brew install llvm@${{ matrix.llvm }} make cmake gcc
  sudo python3 -m pip install --upgrade pip lit
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: mkdir
  run: rm -rf build && mkdir build
  - name: cmake

.github/workflows/enzyme-julia.yml

@@ -28,8 +28,8 @@ jobs:
  - x64
  timeout-minutes: 60 
  steps:
- - uses: actions/checkout@v3
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
+ - uses: actions/checkout@v4
  with:
  repository: 'wsmoses/Enzyme.jl'
  path: ./jl

.github/workflows/enzyme-mlir.yml

@@ -29,14 +29,14 @@ jobs:
  sudo apt-get update
  sudo apt-get install -y binutils ninja-build cmake gcc g++ python3 python3-dev
 
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  with:
  path: 'Enzyme'
 
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  with:
  repository: 'llvm/llvm-project'
- ref: 'bc82cfb38d83f1afeb2c290aa472c2e2e88919cb'
+ ref: '2c9b6c1b36b8185299de083c3058e0c1e7760442'
  path: 'llvm-project'
 
  - name: Get MLIR commit hash
@@ -46,7 +46,7 @@ jobs:
 
  - name: Cache MLIR
  id: cache-mlir
- uses: actions/cache@v3
+ uses: actions/cache@v4
  with:
  path: llvm-project/mlir-build
  key: ${{ matrix.llbuild }}-${{ matrix.os }}-mlir-${{ steps.mlir-commit.outputs.sha_short }}

.github/workflows/format.yml

@@ -11,7 +11,7 @@ jobs:
  runs-on: ubuntu-20.04
 
  steps:
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - uses: DoozyX/clang-format-lint-action@v0.16.2
  with:
  source: 'enzyme/Enzyme enzyme/tools/enzyme-tblgen'

.github/workflows/fortran.yml

@@ -40,7 +40,7 @@ jobs:
  sudo apt-get update && sudo apt-get install -y intel-oneapi-compiler-fortran-${{ matrix.ifx }} intel-oneapi-mpi-${{ matrix.mpi }} intel-oneapi-mpi-devel-${{ matrix.mpi }}
  source /opt/intel/oneapi/setvars.sh
  printenv >> $GITHUB_ENV
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  - name: generate build system
  run: |
  rm -rf build && mkdir build && cd build

.github/workflows/tagger.yml

@@ -10,19 +10,19 @@ jobs:
  name: Enzyme Tag CI
  runs-on: ubuntu-latest
  steps:
- - uses: tibdex/github-app-token@v1
+ - uses: actions/create-github-app-token@v1
  id: generate_token
  with:
  app_id: ${{ secrets.APP_ID }}
  private_key: ${{ secrets.APP_PRIVATE_KEY }}
- repository: JuliaPackaging/Yggdrasil
+ repositories: JuliaPackaging/Yggdrasil
 
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  with:
  repository: 'JuliaPackaging/Yggdrasil'
  path: ygg
 
- - uses: actions/checkout@v3
+ - uses: actions/checkout@v4
  with:
  path: enz
  - name: replace
@@ -37,7 +37,7 @@ jobs:
  git add .
  - name: Create Pull Request
  id: cpr
- uses: peter-evans/create-pull-request@v3
+ uses: peter-evans/create-pull-request@v6
  with:
  path: ygg
  commit-message: "Upgrade enzyme to ${{ github.ref }}"

.gitignore

@@ -9,3 +9,4 @@ enzyme/benchmarks/ReverseMode/*/*.o
 enzyme/benchmarks/ReverseMode/*/*.exe
 enzyme/benchmarks/ReverseMode/*/results.txt
 enzyme/benchmarks/ReverseMode/*/results.json
+.cache

.packaging/build_tarballs.jl

@@ -28,7 +28,7 @@ platforms = expand_cxxstring_abis(supported_platforms(; experimental=true))
 script = raw"""
 cd Enzyme
 
-if [[ "${bb_full_target}" == x86_64-apple-darwin*llvm_version+15.asserts* ]]; then
+if [[ "${bb_full_target}" == x86_64-apple-darwin*llvm_version+15* ]] || [[ "${bb_full_target}" == x86_64-apple-darwin*llvm_version+16* ]]; then
  # LLVM 15 requires macOS SDK 10.14.
  pushd $WORKSPACE/srcdir/MacOSX10.*.sdk
  rm -rf /opt/${target}/${target}/sys-root/System
@@ -54,30 +54,42 @@ cmake -B build-native -S enzyme -GNinja "${NATIVE_CMAKE_FLAGS[@]}"
 
 # Only build blasheaders and tblgen
 ninja -C build-native -j ${nproc} blasheaders enzyme-tblgen
-
 # 2. Cross-compile
 CMAKE_FLAGS=()
 CMAKE_FLAGS+=(-DENZYME_EXTERNAL_SHARED_LIB=ON)
 CMAKE_FLAGS+=(-DBC_LOAD_HEADER=`pwd`/build-native/BCLoad/gsl/blas_headers.h)
+CMAKE_FLAGS+=(-DEnzyme_TABLEGEN=`pwd`/build-native/tools/enzyme-tblgen/enzyme-tblgen)
 CMAKE_FLAGS+=(-DEnzyme_TABLEGEN_EXE=`pwd`/build-native/tools/enzyme-tblgen/enzyme-tblgen)
 CMAKE_FLAGS+=(-DENZYME_CLANG=OFF)
 # RelWithDebInfo for decent performance, with debugability
 CMAKE_FLAGS+=(-DCMAKE_BUILD_TYPE=RelWithDebInfo)
 # Install things into $prefix
 CMAKE_FLAGS+=(-DCMAKE_INSTALL_PREFIX=${prefix})
 # Explicitly use our cmake toolchain file and tell CMake we're cross-compiling
-CMAKE_FLAGS+=(-DCMAKE_TOOLCHAIN_FILE=${CMAKE_TARGET_TOOLCHAIN})
+if [[ "${target}" == *mingw* && "${bb_full_target}" == *llvm_version+16* ]]; then
+ CMAKE_FLAGS+=(-DCMAKE_TOOLCHAIN_FILE=${CMAKE_TARGET_TOOLCHAIN%.*}_clang.cmake)
+else
+ CMAKE_FLAGS+=(-DCMAKE_TOOLCHAIN_FILE=${CMAKE_TARGET_TOOLCHAIN})
+fi
 CMAKE_FLAGS+=(-DCMAKE_CROSSCOMPILING:BOOL=ON)
 # Tell CMake where LLVM is
 CMAKE_FLAGS+=(-DLLVM_DIR="${prefix}/lib/cmake/llvm")
 # Force linking against shared lib
 CMAKE_FLAGS+=(-DLLVM_LINK_LLVM_DYLIB=ON)
 # Build the library
 CMAKE_FLAGS+=(-DBUILD_SHARED_LIBS=ON)
+
+if [[ "${bb_full_target}" == x86_64-apple-darwin*llvm_version+15* ]] || [[ "${bb_full_target}" == x86_64-apple-darwin*llvm_version+16* ]]; then
+if [[ "${target}" == x86_64-apple* ]]; then
+ CMAKE_FLAGS+=(-DCMAKE_OSX_DEPLOYMENT_TARGET:STRING=10.14)
+fi
+else
 if [[ "${target}" == x86_64-apple* ]]; then
  CMAKE_FLAGS+=(-DCMAKE_OSX_DEPLOYMENT_TARGET:STRING=10.12)
 fi
+fi
 
+echo ${CMAKE_FLAGS[@]}
 cmake -B build -S enzyme -GNinja ${CMAKE_FLAGS[@]}
 
 ninja -C build -j ${nproc} install
@@ -113,11 +125,11 @@ for llvm_version in llvm_versions, llvm_assertions in (false, true)
  # We don't build LLVM 15 for i686-linux-musl.
  filter!(p -> !(arch(p) == "i686" && libc(p) == "musl"), platforms)
  end
-
+ 
  for platform in platforms
  augmented_platform = deepcopy(platform)
  augmented_platform[LLVM.platform_name] = LLVM.platform(llvm_version, llvm_assertions)
- gcc_version = version > v"15" ? v"10" : v"8"
+ gcc_version = llvm_version > v"15" ? v"10" : v"8"
  should_build_platform(triplet(augmented_platform)) || continue
  push!(builds, (;
  dependencies, products,

enzyme/BCLoad/BCLoader.cpp

@@ -24,28 +24,30 @@ static inline bool endsWith(llvm::StringRef string, llvm::StringRef suffix) {
 #endif // LLVM_VERSION_MAJOR
 }
 
-bool provideDefinitions(Module &M, std::set<std::string> ignoreFunctions = {}) {
+bool provideDefinitions(Module &M, std::set<std::string> ignoreFunctions,
+ std::vector<std::string> &replaced) {
  std::vector<StringRef> todo;
  bool seen32 = false;
  bool seen64 = false;
  for (auto &F : M) {
  if (!F.empty())
  continue;
+ if (ignoreFunctions.count(F.getName().str()))
+ continue;
  int index = 0;
  for (auto postfix : {"", "_", "_64_"}) {
  std::string str;
  if (strlen(postfix) == 0) {
  str = F.getName().str();
- if (ignoreFunctions.count(str)) continue;
  } else if (endsWith(F.getName(), postfix)) {
  auto blasName =
  F.getName().substr(0, F.getName().size() - strlen(postfix)).str();
- if (ignoreFunctions.count(blasName)) continue;
  str = "cblas_" + blasName;
  }
 
  auto found = EnzymeBlasBC.find(str);
  if (found != EnzymeBlasBC.end()) {
+ replaced.push_back(F.getName().str());
  todo.push_back(found->second);
  if (index == 1)
  seen32 = true;
@@ -81,13 +83,24 @@ bool provideDefinitions(Module &M, std::set<std::string> ignoreFunctions = {}) {
  });
 #endif
 
- if (!BC)
+ if (!BC) {
  Err.print("bcloader", llvm::errs());
+ continue;
+ }
  assert(BC);
  SmallVector<std::string, 1> toReplace;
  for (auto &F : *BC) {
  if (F.empty())
  continue;
+ if (ignoreFunctions.count(F.getName().str())) {
+ F.dropAllReferences();
+#if LLVM_VERSION_MAJOR >= 16
+ F.erase(F.begin(), F.end());
+#else
+ F.getBasicBlockList().erase(F.begin(), F.end());
+#endif
+ continue;
+ }
  toReplace.push_back(F.getName().str());
  }
  BC->setTargetTriple("");
@@ -106,12 +119,29 @@ bool provideDefinitions(Module &M, std::set<std::string> ignoreFunctions = {}) {
 
 extern "C" {
 uint8_t EnzymeBitcodeReplacement(LLVMModuleRef M, char **FncsNamesToIgnore,
- size_t numFncNames) {
+ size_t numFncNames, const char ***foundP,
+ size_t *foundLen) {
  std::set<std::string> ignoreFunctions = {};
  for (size_t i = 0; i < numFncNames; i++) {
  ignoreFunctions.insert(std::string(FncsNamesToIgnore[i]));
  }
- return provideDefinitions(*unwrap(M), ignoreFunctions);
+ std::vector<std::string> replaced;
+ auto res = provideDefinitions(*unwrap(M), ignoreFunctions, replaced);
+
+ const char **found = nullptr;
+ if (replaced.size()) {
+ found = (const char **)malloc(replaced.size() * sizeof(const char **));
+ for (size_t i = 0; i < replaced.size(); i++) {
+ char *data = (char *)malloc(replaced[i].size() + 1);
+ memcpy(data, replaced[i].data(), replaced[i].size());
+ data[replaced[i].size()] = 0;
+ found[i] = data;
+ }
+ }
+ *foundP = found;
+ *foundLen = replaced.size();
+
+ return res;
 }
 }
 
@@ -121,7 +151,10 @@ class BCLoader final : public ModulePass {
  static char ID;
  BCLoader() : ModulePass(ID) {}
 
- bool runOnModule(Module &M) override { return provideDefinitions(M, {}); }
+ bool runOnModule(Module &M) override {
+ std::vector<std::string> replaced;
+ return provideDefinitions(M, {}, replaced);
+ }
 };
 } // namespace
 

enzyme/BCLoad/CMakeLists.txt

@@ -4,10 +4,10 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(BC_LOAD_FLAGS "" CACHE STRING "")
 set(BC_LOAD_HEADER "" CACHE STRING "")
 
-if (${LLVM_VERSION_MAJOR} LESS 15)
- set(BC_LOAD_FLAGS2 "${BC_LOAD_FLAGS}")
-else()
+if (${LLVM_VERSION_MAJOR} EQUAL 15 OR ${LLVM_VERSION_MAJOR} EQUAL 16)
  set(BC_LOAD_FLAGS2 "${BC_LOAD_FLAGS} -Xclang -no-opaque-pointers")
+else()
+ set(BC_LOAD_FLAGS2 "${BC_LOAD_FLAGS}")
 endif()
 
 if (APPLE)