LTO support

.github/workflows/benchmark_compare.yml

@@ -37,7 +37,7 @@ jobs:
     steps:
       - name: Get PR details
         id: pr
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             const pullRequestNumber = context.eventName === 'workflow_dispatch'
@@ -60,7 +60,7 @@ jobs:
             };
 
       - name: Create status check
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             const prData = ${{ steps.pr.outputs.result }};
@@ -75,7 +75,7 @@ jobs:
             });
 
       - name: Checkout base repository
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
         with:
           fetch-depth: 0
 
@@ -96,7 +96,7 @@ jobs:
           echo "timestamp=$(date -u +'%Y-%m-%d-%H;%M;%S')" >> $GITHUB_OUTPUT
 
       - name: Setup ccache cache files
-        uses: actions/cache@v5
+        uses: actions/cache@v6
         with:
           path: ${{github.workspace}}/build/.ccache
           key: benchmark-ccache-${{ steps.ccache_cache_timestamp.outputs.timestamp }}
@@ -114,7 +114,7 @@ jobs:
 
       - name: Upload Results
         if: always()
-        uses: actions/upload-artifact@v6
+        uses: actions/upload-artifact@v7
         with:
           name: benchmark-data-${{ github.run_id }}
           path: |
@@ -125,7 +125,7 @@ jobs:
 
       - name: Post Results
         if: success()
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             const fs = require('fs');
@@ -156,7 +156,7 @@ jobs:
 
       - name: Update status check - Success
         if: success()
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             const prData = ${{ steps.pr.outputs.result }};
@@ -172,7 +172,7 @@ jobs:
 
       - name: Notify on Failure
         if: failure()
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             const prData = ${{ steps.pr.outputs.result }};

.github/workflows/cmake.yml

@@ -63,14 +63,14 @@ jobs:
         -DSEQUANT_PYTHON=${{ !matrix.valgrind && !matrix.sanitize }}
         -DSEQUANT_ASSERT_BEHAVIOR=${{ matrix.valgrind && 'IGNORE' || 'THROW' }}
     steps:
-    - uses: actions/checkout@v6
+    - uses: actions/checkout@v7
 
     - name: free up disk space on Ubuntu runner
       if: ${{ matrix.os == 'ubuntu-24.04' }}
       uses: jlumbroso/free-disk-space@main
 
     - name: Setup Python
-      uses: actions/setup-python@v6
+      uses: actions/setup-python@v7
       with:
         python-version: '3.x'
 
@@ -106,7 +106,7 @@ jobs:
       run: echo "timestamp=$(date -u +'%Y-%m-%d-%H-%M-%S')" >> $GITHUB_OUTPUT
 
     - name: Setup ccache cache files
-      uses: actions/cache@v5
+      uses: actions/cache@v6
       with:
         path: ${{github.workspace}}/build/.ccache
         key: ${{ matrix.config.name }}-ccache-${{ steps.ccache_cache_timestamp.outputs.timestamp }}
@@ -121,7 +121,7 @@ jobs:
       # Note the current convention is to use the -S and -B options here to specify source 
       # and build directories, but this is only available with CMake 3.13 and higher.  
       # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE $BUILD_CONFIG
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE $BUILD_CONFIG -DSEQUANT_QUICK_BENCHMARKS=ON
 
 
     - name: Build
@@ -136,6 +136,12 @@ jobs:
       shell: bash
       run: ctest --output-on-failure -R "^sequant"
 
+    - name: Ensure benchmarks not broken
+      if: ${{ !matrix.valgrind && !matrix.sanitize }}
+      working-directory: ${{github.workspace}}/build
+      shell: bash
+      run: ./benchmarks/sequant_benchmarks --benchmark_dry_run=true
+
     - name: Test (+ Valgrind)
       if: ${{ matrix.valgrind }}
       working-directory: ${{github.workspace}}/build

.github/workflows/docs.yml

@@ -16,8 +16,8 @@ jobs:
         -DSEQUANT_BUILD_DOCS=ON
         -DSEQUANT_BUILD_DOCS_API_BREATHEEXHALE=ON
     steps:
-      - uses: actions/checkout@v6
-      - uses: actions/setup-python@v6
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v7
         with:
           python-version: '>=3.9' # Breathe needs 3.9 or later
           cache: 'pip'

.github/workflows/formatting_check.yml

@@ -7,7 +7,7 @@ jobs:
     name: "Check formatting"
     runs-on: "ubuntu-latest"
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
 
       - name: Install clang-format-17
         run: |

CMakeLists.txt

@@ -95,6 +95,7 @@ macro(sequant_add_library _module)
     get_target_property(_type SeQuant-${_module} TYPE)
     if (NOT _type STREQUAL "INTERFACE_LIBRARY")
         target_set_warning_flags(SeQuant-${_module})
+        target_set_optimization_flags(SeQuant-${_module})
     endif()
 endmacro()
 
@@ -272,6 +273,7 @@ add_library(SeQuant-bliss
 set_target_properties(SeQuant-bliss PROPERTIES EXPORT_NAME bliss)
 set_target_properties(SeQuant-bliss PROPERTIES CXX_SCAN_FOR_MODULES OFF)
 add_library(SeQuant::bliss ALIAS SeQuant-bliss)
+target_set_optimization_flags(SeQuant-bliss)
 target_link_libraries(SeQuant-bliss PUBLIC range-v3::range-v3 Boost::headers)
 if (SEQUANT_USE_SYSTEM_BOOST_HASH)
     target_compile_definitions(SeQuant-bliss PUBLIC SEQUANT_USE_SYSTEM_BOOST_HASH=1)

SeQuant/core/eval/eval.hpp

@@ -557,7 +557,8 @@ ResultPtr evaluate(Node const& node,  //
     bool checked;
     Stage stage = Stage::Enter;
     bool store_after = false;
-    ResultPtr left, right;
+    ResultPtr left = {};
+    ResultPtr right = {};
   };
 
   // Finalize a freshly computed Phase-B result: if this Checked node needs

SeQuant/domain/mbpt/op.cpp

@@ -580,14 +580,14 @@ ExprPtr OpMaker<S>::operator()(
   if (!dep && csv) {
     if (opclass == OpClass::Ex) {
       if constexpr (assert_enabled()) {
-        for (auto&& s : cre_spaces_) {
+        for ([[maybe_unused]] const auto& s : cre_spaces_) {
           SEQUANT_ASSERT(isr->contains_unoccupied(s));
         }
       }
       dep = UseDepIdx::Bra;
     } else if (opclass == OpClass::Deex) {
       if constexpr (assert_enabled()) {
-        for (auto&& s : ann_spaces_) {
+        for ([[maybe_unused]] const auto& s : ann_spaces_) {
           SEQUANT_ASSERT(isr->contains_unoccupied(s));
         }
       }

benchmarks/CMakeLists.txt

@@ -11,8 +11,23 @@ add_executable(sequant_benchmarks
         "wick.cpp"
 )
 
+target_set_warning_flags(sequant_benchmarks)
+target_set_optimization_flags(sequant_benchmarks)
+
 set_target_properties(sequant_benchmarks PROPERTIES CXX_SCAN_FOR_MODULES OFF)
 
+option(SEQUANT_QUICK_BENCHMARKS "Whether to exclude computational expensive benchmarks" OFF)
+
+# Running benchmarks in Debug mode doesn't make a lot of sense but in case someone does,
+# we at least want to exclude benchmarks that are likely to take forever in this setup
+if (CMAKE_BUILD_TYPE STREQUAL "Debug" OR SEQUANT_QUICK_BENCHMARKS)
+	target_compile_definitions(sequant_benchmarks
+		PRIVATE
+			SEQUANT_BENCH_MAX_CC_RANK=3
+			SEQUANT_BENCH_MAX_WICK_INPUTS=2
+	)
+endif()
+
 # The benchmark implementations contain a bunch of same-name symbol definition that
 # ought to be local to the individual translation unit. Hence, we must not combine
 # all of them into a single translation unit.

benchmarks/coupled_cluster.cpp

@@ -2,7 +2,11 @@
 
 #include <SeQuant/domain/mbpt/models/cc.hpp>
 
+#ifdef SEQUANT_BENCH_MAX_CC_RANK
+static constexpr std::size_t maxRank = SEQUANT_BENCH_MAX_CC_RANK;
+#else
 static constexpr std::size_t maxRank = 10;
+#endif
 
 using namespace sequant;
 using namespace sequant::mbpt;

benchmarks/main.cpp

@@ -2,6 +2,7 @@
 
 #include <SeQuant/core/context.hpp>
 #include <SeQuant/core/runtime.hpp>
+#include <SeQuant/domain/mbpt/context.hpp>
 #include <SeQuant/domain/mbpt/convention.hpp>
 
 using namespace sequant;
@@ -17,6 +18,9 @@ int main(int argc, char *argv[]) {
                                .vacuum = Vacuum::SingleProduct});
   set_default_context(fermi_ctx);
 
+  mbpt::Context mbpt_ctx({.op_registry_ptr = mbpt::make_minimal_registry()});
+  mbpt::set_default_mbpt_context(std::move(mbpt_ctx));
+
   Context bose_einstein_ctx = Context(
       {.index_space_registry_shared_ptr = idxreg, .vacuum = Vacuum::Physical});
 

benchmarks/wick.cpp

@@ -13,7 +13,11 @@
 using namespace sequant;
 using namespace sequant::mbpt;
 
+#ifdef SEQUANT_BENCH_MAX_WICK_INPUTS
+static constexpr std::size_t nInputs = SEQUANT_BENCH_MAX_WICK_INPUTS;
+#else
 static constexpr std::size_t nInputs = 5;
+#endif
 
 template <Statistics stats>
 ExprPtr get_op_sequence(std::size_t i) {
@@ -147,8 +151,9 @@ VacAvPair get_mbpt_expr(std::size_t i) {
 }
 
 static void mbpt_vac_av(benchmark::State &state, bool csv) {
-  auto ctx = sequant::mbpt::set_scoped_default_mbpt_context(
-      mbpt::Context({.csv = csv ? CSV::Yes : CSV::No}));
+  auto base = mbpt::get_default_mbpt_context();
+  base.set(csv ? CSV::Yes : CSV::No);
+  auto ctx = sequant::mbpt::set_scoped_default_mbpt_context(base);
 
   VacAvPair input = get_mbpt_expr(state.range(0));
 

cmake/compiler.cmake

@@ -60,3 +60,66 @@ function(target_set_warning_flags TARGET)
         target_compile_options("${TARGET}" PRIVATE "-Wno-unused-lambda-capture")
     endif()
 endfunction()
+
+
+
+function(target_set_optimization_flags TARGET)
+	if (CMAKE_BUILD_TYPE STREQUAL "Debug")
+		return()
+	endif()
+
+	get_target_property(TARGET_TYPE "${TARGET}" TYPE)
+
+	if (TARGET_TYPE STREQUAL "INTERFACE_LIBRARY")
+		message(WARNING "target_set_optimization_flags is not intended to be used on interface targets")
+		return()
+	endif()
+
+	include(CheckCXXCompilerFlag)
+	include(CheckIPOSupported)
+
+	check_ipo_supported(RESULT CMAKE_SUPPORTS_COMPILER_LTO LANGUAGES CXX)
+
+	set(CMAKE_TRY_COMPILE_TARGET_TYPE "STATIC_LIBRARY")
+	check_cxx_compiler_flag("-flto" SEQUANT_LTO_FLAG_SUPPORTED)
+	check_cxx_compiler_flag("-flto=auto" SEQUANT_LTO_AUTO_SUPPORTED)
+	check_cxx_compiler_flag("-flto;-ffat-lto-objects" SEQUANT_FAT_LTO_FLAG_SUPPORTED)
+
+	if (DEFINED SEQUANT_LTO)
+		# Always honor explicit user choice
+		set(ENABLE_LTO ${SEQUANT_LTO})
+	elseif(TARGET_TYPE STREQUAL "STATIC_LIBRARY" OR TARGET_TYPE STREQUAL "OBJECT_LIBRARY")
+		# For static/object libraries we only want to enable LTO by default, if we can create
+		# "fat" object files. Those can still be linked without LTO and hence shouldn't
+		# break any downstream use.
+		set(ENABLE_LTO ${SEQUANT_FAT_LTO_FLAG_SUPPORTED})
+	elseif(SEQUANT_LTO_FLAG_SUPPORTED OR CMAKE_SUPPORTS_COMPILER_LTO)
+		# Anything but static/object libraries is also linked by us and
+		# hence enabling LTO doesn't affect downstream compatibility
+		set(ENABLE_LTO ON)
+	endif()
+
+	if (ENABLE_LTO)
+		if (SEQUANT_LTO_FLAG_SUPPORTED)
+			# We prefer to manually set the LTO flag(s) rather than CMake doing it for us
+			# due to https://gitlab.kitware.com/cmake/cmake/-/work_items/23136
+			# On some compilers, the thin LTO type requested by CMake is incompatible
+			# with explicitly asking for fat LTO object files.
+			# Besides, it seems like full LTO achieves quite a bit better optimizations
+			# with Clang.
+			if (SEQUANT_LTO_AUTO_SUPPORTED)
+				target_compile_options("${TARGET}" PRIVATE -flto=auto)
+				target_link_options("${TARGET}" PRIVATE -flto=auto)
+			else()
+				target_compile_options("${TARGET}" PRIVATE -flto)
+				target_link_options("${TARGET}" PRIVATE -flto)
+			endif()
+
+			if (SEQUANT_FAT_LTO_FLAG_SUPPORTED)
+				target_compile_options("${TARGET}" PRIVATE -ffat-lto-objects)
+			endif()
+		else()
+			set_target_properties("${TARGET}" PROPERTIES INTERPROCEDURAL_OPTIMIZATION ON)
+		endif()
+	endif()
+endfunction()

doc/user/getting_started/installing.rst

@@ -110,6 +110,9 @@ Useful CMake Variables
      - ``ABORT`` in ``Debug`` mode, ``IGNORE`` otherwise
      - Controls how assertions within SeQuant's code are handled. Valid options are ``ABORT``, ``THROW`` and ``IGNORE``. The latter disables
        assertions, whereas the former keep them active and either abort the program or throw an exception on violation respectively.
+   * - SEQUANT_LTO
+     - `ON` if the compiler supportes "fat" LTO objects, `OFF` otherwise
+     - Controls whether SeQuant will be built with link-time optimizations (LTO)
 
 
 Configuring and Building

external/versions.cmake

@@ -11,12 +11,12 @@ set(SEQUANT_TRACKED_LIBPERM_TAG cada3e185549896203cf4d0c7f26ea22c7de428f)
 
 set(SEQUANT_TRACKED_POLYMORPHICVARIANT_TAG 010c69786104c07c5faccffe0e99f99de5a69fd8)
 
-set(SEQUANT_TRACKED_UTFCPP_TAG v4.0.6)
+set(SEQUANT_TRACKED_UTFCPP_TAG v4.1.1)
 
-set(SEQUANT_TRACKED_CLI11_TAG v2.5.0)
+set(SEQUANT_TRACKED_CLI11_TAG v2.7.2)
 set(SEQUANT_OLDEST_CLI11_VERSION 2)
 
-set(SEQUANT_TRACKED_SPDLOG_TAG v1.15.3)
+set(SEQUANT_TRACKED_SPDLOG_TAG v1.17.0)
 
 set(SEQUANT_TRACKED_JSON_TAG v3.12.0)
 set(SEQUANT_OLDEST_JSON_VERSION 3)
@@ -27,13 +27,13 @@ set(SEQUANT_OLDEST_JSON_VERSION 3)
 #   Recent stdlibs have removed this class when using C++17 (e.g. Apple Clang 15)
 set(SEQUANT_OLDEST_BOOST_VERSION 1.81)
 
-set(SEQUANT_TRACKED_CATCH2_TAG v3.9.1)
+set(SEQUANT_TRACKED_CATCH2_TAG v3.15.3)
 set(SEQUANT_OLDEST_CATCH2_VERSION 3.3)
 
-set(SEQUANT_TRACKED_GOOGLEBENCHMARK_TAG v1.9.4)
+set(SEQUANT_TRACKED_GOOGLEBENCHMARK_TAG v1.9.5)
 set(SEQUANT_OLDEST_GOOGLEBENCHMARK_VERSION 1.9.3)
 
-set(SEQUANT_TRACKED_PYBIND11_TAG v3.0.1)
+set(SEQUANT_TRACKED_PYBIND11_TAG v3.1.0)
 set(SEQUANT_OLDEST_PYBIND11_VERSION 3)
 
 # oldest Doxygen we can tolerate ... doxygen-awesome-css requires doxygen 1.9.1 - 1.9.4 or 1.9.6 - 1.14.0, so just use 1.9.6 as the minimum

python/CMakeLists.txt

@@ -13,6 +13,7 @@ endif()
 pybind11_add_module(python-sequant MODULE src/sequant/_sequant.cc)
 
 target_link_libraries(python-sequant PRIVATE SeQuant)
+target_set_optimization_flags(python-sequant)
 target_include_directories(python-sequant PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/include")
 
 set_target_properties(

tests/integration/CMakeLists.txt

@@ -56,6 +56,7 @@ foreach(current IN LISTS TEST_CASES)
     endif()
 
     target_set_warning_flags("${test_name}")
+    target_set_optimization_flags("${test_name}")
 
     set(variant_names "")