Bump LLVM from 12 to 13

.gitmodules

@@ -30,9 +30,6 @@
 [submodule "contrib/re2"]
 	path = contrib/re2
 	url = https://github.com/google/re2.git
-[submodule "contrib/llvm"]
-	path = contrib/llvm
-	url = https://github.com/ClickHouse/llvm
 [submodule "contrib/mariadb-connector-c"]
 	path = contrib/mariadb-connector-c
 	url = https://github.com/ClickHouse/mariadb-connector-c.git
@@ -284,3 +281,6 @@
 [submodule "contrib/c-ares"]
 	path = contrib/c-ares
 	url = https://github.com/ClickHouse/c-ares
+[submodule "contrib/llvm-project"]
+	path = contrib/llvm-project
+	url = https://github.com/ClickHouse/llvm-project.git

contrib/CMakeLists.txt

@@ -107,7 +107,7 @@ if (ENABLE_TESTS)
     add_contrib (googletest-cmake googletest)
 endif()
 
-add_contrib (llvm-cmake llvm)
+add_contrib (llvm-project-cmake llvm-project)
 add_contrib (libxml2-cmake libxml2)
 add_contrib (aws-s3-cmake
     aws

contrib/llvm

@@ -1 +0,0 @@
-Subproject commit 0db5bf5bd2452cd8f1283a1fcdc04845af705bfc

contrib/llvm-cmake/CMakeLists.txt

@@ -1,112 +0,0 @@
-if (APPLE OR NOT ARCH_AMD64 OR SANITIZE STREQUAL "undefined")
-   set (ENABLE_EMBEDDED_COMPILER_DEFAULT OFF)
-else()
-   set (ENABLE_EMBEDDED_COMPILER_DEFAULT ON)
-endif()
-
-option (ENABLE_EMBEDDED_COMPILER "Enable support for 'compile_expressions' option for query execution" ${ENABLE_EMBEDDED_COMPILER_DEFAULT})
-
-if (NOT ENABLE_EMBEDDED_COMPILER)
-    message(STATUS "Not using LLVM")
-    return()
-endif()
-
-set (LLVM_FOUND 1)
-set (LLVM_VERSION "12.0.0bundled")
-set (LLVM_INCLUDE_DIRS
-    "${ClickHouse_SOURCE_DIR}/contrib/llvm/llvm/include"
-    "${ClickHouse_BINARY_DIR}/contrib/llvm/llvm/include"
-)
-set (LLVM_LIBRARY_DIRS "${ClickHouse_BINARY_DIR}/contrib/llvm/llvm")
-
-# This list was generated by listing all LLVM libraries, compiling the binary and removing all libraries while it still compiles.
-set (REQUIRED_LLVM_LIBRARIES
-    LLVMExecutionEngine
-    LLVMRuntimeDyld
-    LLVMAsmPrinter
-    LLVMDebugInfoDWARF
-    LLVMGlobalISel
-    LLVMSelectionDAG
-    LLVMMCDisassembler
-    LLVMPasses
-    LLVMCodeGen
-    LLVMipo
-    LLVMBitWriter
-    LLVMInstrumentation
-    LLVMScalarOpts
-    LLVMAggressiveInstCombine
-    LLVMInstCombine
-    LLVMVectorize
-    LLVMTransformUtils
-    LLVMTarget
-    LLVMAnalysis
-    LLVMProfileData
-    LLVMObject
-    LLVMBitReader
-    LLVMCore
-    LLVMRemarks
-    LLVMBitstreamReader
-    LLVMMCParser
-    LLVMMC
-    LLVMBinaryFormat
-    LLVMDebugInfoCodeView
-    LLVMSupport
-    LLVMDemangle
-)
-
-if (ARCH_AMD64)
-    list(APPEND REQUIRED_LLVM_LIBRARIES LLVMX86Info LLVMX86Desc LLVMX86CodeGen)
-elseif (ARCH_AARCH64)
-    list(APPEND REQUIRED_LLVM_LIBRARIES LLVMAArch64Info LLVMAArch64Desc LLVMAArch64CodeGen)
-endif ()
-
-#function(llvm_libs_all REQUIRED_LLVM_LIBRARIES)
-#    llvm_map_components_to_libnames (result all)
-#    if (USE_STATIC_LIBRARIES OR NOT "LLVM" IN_LIST result)
-#        list (REMOVE_ITEM result "LTO" "LLVM")
-#    else()
-#        set (result "LLVM")
-#    endif ()
-#    list (APPEND result ${CMAKE_DL_LIBS} ch_contrib::zlib)
-#    set (${REQUIRED_LLVM_LIBRARIES} ${result} PARENT_SCOPE)
-#endfunction()
-
-message (STATUS "LLVM include Directory: ${LLVM_INCLUDE_DIRS}")
-message (STATUS "LLVM library Directory: ${LLVM_LIBRARY_DIRS}")
-message (STATUS "LLVM C++ compiler flags: ${LLVM_CXXFLAGS}")
-
-# ld: unknown option: --color-diagnostics
-set (LINKER_SUPPORTS_COLOR_DIAGNOSTICS 0 CACHE INTERNAL "")
-
-# Do not adjust RPATH in llvm, since then it will not be able to find libcxx/libcxxabi/libunwind
-set (CMAKE_INSTALL_RPATH "ON")
-set (LLVM_COMPILER_CHECKED 1 CACHE INTERNAL "")
-set (LLVM_ENABLE_EH 1 CACHE INTERNAL "")
-set (LLVM_ENABLE_RTTI 1 CACHE INTERNAL "")
-set (LLVM_ENABLE_PIC 0 CACHE INTERNAL "")
-set (LLVM_TARGETS_TO_BUILD "X86;AArch64" CACHE STRING "")
-
-# Need to use C++17 since the compilation is not possible with C++20 currently, due to ambiguous operator != etc.
-# LLVM project will set its default value for the -std=... but our global setting from CMake will override it.
-set (CMAKE_CXX_STANDARD 17)
-
-set (LLVM_SOURCE_DIR "${ClickHouse_SOURCE_DIR}/contrib/llvm/llvm")
-set (LLVM_BINARY_DIR "${ClickHouse_BINARY_DIR}/contrib/llvm/llvm")
-add_subdirectory ("${LLVM_SOURCE_DIR}" "${LLVM_BINARY_DIR}")
-set_directory_properties (PROPERTIES
-   # due to llvm crosscompile cmake does not know how to clean it, and on clean
-   # will lead to the following error:
-   #
-   #   ninja: error: remove(contrib/llvm/llvm/NATIVE): Directory not empty
-   #
-   ADDITIONAL_CLEAN_FILES "${LLVM_BINARY_DIR}"
-   # llvm's cmake configuring this file only when cmake runs,
-   # and after clean cmake will not know that it should re-run,
-   # add explicitly depends from llvm-config.h
-   CMAKE_CONFIGURE_DEPENDS "${LLVM_BINARY_DIR}/include/llvm/Config/llvm-config.h"
-)
-
-add_library (_llvm INTERFACE)
-target_link_libraries (_llvm INTERFACE ${REQUIRED_LLVM_LIBRARIES})
-target_include_directories (_llvm SYSTEM BEFORE INTERFACE ${LLVM_INCLUDE_DIRS})
-add_library(ch_contrib::llvm ALIAS _llvm)

contrib/llvm-project

@@ -0,0 +1 @@
+Subproject commit 71e5e5f48761c6ae1b17c1e5afe1d1dcfdbafdec

contrib/llvm-project-cmake/CMakeLists.txt

@@ -0,0 +1,122 @@
+if (APPLE OR NOT ARCH_AMD64 OR SANITIZE STREQUAL "undefined" OR NOT USE_STATIC_LIBRARIES)
+   set (ENABLE_EMBEDDED_COMPILER_DEFAULT OFF)
+else()
+   set (ENABLE_EMBEDDED_COMPILER_DEFAULT ON)
+endif()
+
+option (ENABLE_EMBEDDED_COMPILER "Enable support for 'compile_expressions' option for query execution" ${ENABLE_EMBEDDED_COMPILER_DEFAULT})
+
+if (NOT ENABLE_EMBEDDED_COMPILER)
+    message(STATUS "Not using LLVM")
+    return()
+endif()
+
+# TODO: Enable shared library build
+# TODO: Enable compilation on AArch64
+
+set (LLVM_VERSION "13.0.0bundled")
+set (LLVM_INCLUDE_DIRS
+    "${ClickHouse_SOURCE_DIR}/contrib/llvm-project/llvm/include"
+    "${ClickHouse_BINARY_DIR}/contrib/llvm-project/llvm/include"
+)
+set (LLVM_LIBRARY_DIRS "${ClickHouse_BINARY_DIR}/contrib/llvm-project/llvm")
+
+# This list was generated by listing all LLVM libraries, compiling the binary and removing all libraries while it still compiles.
+set (REQUIRED_LLVM_LIBRARIES
+    LLVMExecutionEngine
+    LLVMRuntimeDyld
+    LLVMAsmPrinter
+    LLVMDebugInfoDWARF
+    LLVMGlobalISel
+    LLVMSelectionDAG
+    LLVMMCDisassembler
+    LLVMPasses
+    LLVMCodeGen
+    LLVMipo
+    LLVMBitWriter
+    LLVMInstrumentation
+    LLVMScalarOpts
+    LLVMAggressiveInstCombine
+    LLVMInstCombine
+    LLVMVectorize
+    LLVMTransformUtils
+    LLVMTarget
+    LLVMAnalysis
+    LLVMProfileData
+    LLVMObject
+    LLVMBitReader
+    LLVMCore
+    LLVMRemarks
+    LLVMBitstreamReader
+    LLVMMCParser
+    LLVMMC
+    LLVMBinaryFormat
+    LLVMDebugInfoCodeView
+    LLVMSupport
+    LLVMDemangle
+)
+
+# if (ARCH_AMD64)
+    list(APPEND REQUIRED_LLVM_LIBRARIES LLVMX86Info LLVMX86Desc LLVMX86CodeGen)
+# elseif (ARCH_AARCH64)
+#     list(APPEND REQUIRED_LLVM_LIBRARIES LLVMAArch64Info LLVMAArch64Desc LLVMAArch64CodeGen)
+# endif ()
+
+# ld: unknown option: --color-diagnostics
+# set (LINKER_SUPPORTS_COLOR_DIAGNOSTICS 0 CACHE INTERNAL "")
+
+set (CMAKE_INSTALL_RPATH "ON") # Do not adjust RPATH in llvm, since then it will not be able to find libcxx/libcxxabi/libunwind
+set (LLVM_COMPILER_CHECKED 1 CACHE INTERNAL "") # Skip internal compiler selection
+set (LLVM_ENABLE_EH 1 CACHE INTERNAL "") # With exception handling
+set (LLVM_ENABLE_RTTI 1 CACHE INTERNAL "")
+set (LLVM_ENABLE_PIC 0 CACHE INTERNAL "")
+set (LLVM_TARGETS_TO_BUILD "X86" CACHE STRING "") # for x86 + ARM: "X86;AArch64"
+
+# Omit unnecessary stuff (just the options which are ON by default)
+set(LLVM_ENABLE_BACKTRACES 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_CRASH_OVERRIDES 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_TERMINFO 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_LIBXML2 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_LIBEDIT 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_LIBPFM 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_ZLIB 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_Z3_SOLVER 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_TOOLS 0 CACHE INTERNAL "")
+set(LLVM_BUILD_TOOLS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_UTILS 0 CACHE INTERNAL "")
+set(LLVM_BUILD_UTILS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_RUNTIMES 0 CACHE INTERNAL "")
+set(LLVM_BUILD_RUNTIMES 0 CACHE INTERNAL "")
+set(LLVM_BUILD_RUNTIME 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_EXAMPLES 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_TESTS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_GO_TESTS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_BENCHMARKS 0 CACHE INTERNAL "")
+set(LLVM_INCLUDE_DOCS 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_OCAMLDOC 0 CACHE INTERNAL "")
+set(LLVM_ENABLE_BINDINGS 0 CACHE INTERNAL "")
+
+# C++20 is currently not supported due to ambiguous operator != etc.
+set (CMAKE_CXX_STANDARD 17)
+
+set (LLVM_SOURCE_DIR "${ClickHouse_SOURCE_DIR}/contrib/llvm-project/llvm")
+set (LLVM_BINARY_DIR "${ClickHouse_BINARY_DIR}/contrib/llvm-project/llvm")
+add_subdirectory ("${LLVM_SOURCE_DIR}" "${LLVM_BINARY_DIR}")
+
+set_directory_properties (PROPERTIES
+   # due to llvm crosscompile cmake does not know how to clean it, and on clean
+   # will lead to the following error:
+   #
+   #   ninja: error: remove(contrib/llvm/llvm/NATIVE): Directory not empty
+   #
+   ADDITIONAL_CLEAN_FILES "${LLVM_BINARY_DIR}"
+   # llvm's cmake configuring this file only when cmake runs,
+   # and after clean cmake will not know that it should re-run,
+   # add explicitly depends from llvm-config.h
+   CMAKE_CONFIGURE_DEPENDS "${LLVM_BINARY_DIR}/include/llvm/Config/llvm-config.h"
+)
+
+add_library (_llvm INTERFACE)
+target_link_libraries (_llvm INTERFACE ${REQUIRED_LLVM_LIBRARIES})
+target_include_directories (_llvm SYSTEM BEFORE INTERFACE ${LLVM_INCLUDE_DIRS})
+add_library(ch_contrib::llvm ALIAS _llvm)

src/AggregateFunctions/AggregateFunctionNull.h

@@ -235,8 +235,15 @@ public:
 
         if constexpr (result_is_nullable)
         {
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
             auto * aggregate_data_is_null_dst_value = b.CreateLoad(aggregate_data_dst_ptr);
             auto * aggregate_data_is_null_src_value = b.CreateLoad(aggregate_data_src_ptr);
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
 
             auto * is_src_null = nativeBoolCast(b, std::make_shared<DataTypeUInt8>(), aggregate_data_is_null_src_value);
             auto * is_null_result_value = b.CreateSelect(is_src_null, llvm::ConstantInt::get(b.getInt8Ty(), 1), aggregate_data_is_null_dst_value);

src/Interpreters/JIT/compileFunction.cpp

@@ -842,7 +842,14 @@ CompiledSortDescriptionFunction compileSortDescription(
             auto * lhs_column_data = b.CreatePointerCast(b.CreateExtractValue(lhs_column, {0}), column_native_type_pointer);
             auto * lhs_column_null_data = column_type_is_nullable ? b.CreateExtractValue(lhs_column, {1}) : nullptr;
 
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
             llvm::Value * lhs_value = b.CreateLoad(b.CreateInBoundsGEP(nullptr, lhs_column_data, lhs_index_arg));
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
 
             if (lhs_column_null_data)
             {
@@ -856,7 +863,14 @@ CompiledSortDescriptionFunction compileSortDescription(
             auto * rhs_column_data = b.CreatePointerCast(b.CreateExtractValue(rhs_column, {0}), column_native_type_pointer);
             auto * rhs_column_null_data = column_type_is_nullable ? b.CreateExtractValue(rhs_column, {1}) : nullptr;
 
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
             llvm::Value * rhs_value = b.CreateLoad(b.CreateInBoundsGEP(nullptr, rhs_column_data, rhs_index_arg));
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
             if (rhs_column_null_data)
             {
                 auto * is_null_value_pointer = b.CreateInBoundsGEP(nullptr, rhs_column_null_data, rhs_index_arg);