[js] start js bindings

.github/workflows/build_llvm.yml

@@ -19,6 +19,7 @@ on:
       - ".github/workflows/build_llvm.yml"
       - "build_tools/cmake/llvm_cache.cmake"
       - "build_tools/cmake/llvm_wasm_cache.cmake"
+      - "build_tools/cmake/llvm_wasm_project_include.cmake"
       - "build_tools/build_llvm.sh"
       - "build_tools/build_llvm_wasi.sh"
       - "third_party/llvm-project"
@@ -28,6 +29,7 @@ on:
     paths:
       - "build_tools/cmake/llvm_cache.cmake"
       - "build_tools/cmake/llvm_wasm_cache.cmake"
+      - "build_tools/cmake/llvm_wasm_project_include.cmake"
       - "build_tools/build_llvm.sh"
       - "build_tools/build_llvm_wasi.sh"
       - "third_party/llvm-project"

build_tools/build_llvm_wasi.sh

@@ -32,20 +32,22 @@ echo "*********************** BUILDING LLVM *********************************"
 # https://stackoverflow.com/a/75596433/9045206
 sed -i.bak 's/CMAKE_EXECUTABLE_SUFFIX ".js"/CMAKE_EXECUTABLE_SUFFIX ".html"/g' "$EMSDK/upstream/emscripten/cmake/Modules/Platform/Emscripten.cmake"
 
+# all emscripten settings here https://github.com/emscripten-core/emscripten/blob/main/src/settings.js
+# wasm-ld exports https://lld.llvm.org/WebAssembly.html#exports
+# -sLINKABLE=1 combined with CMAKE_CXX_VISIBILITY_PRESET=default will cause all symbols to be exported
+# -sEXPORT_ALL=1 put all of the exported function in the wasm on the JS module
 cmake_options=(
   -GNinja
   -S "${LLVM_SOURCE_DIR}/llvm"
   -B "${LLVM_BUILD_DIR}"
-  # optimize for size
-  -DCMAKE_C_FLAGS="-Os"
-  -DCMAKE_CXX_FLAGS="-Os"
   -DCMAKE_BUILD_TYPE=Release
-  -DCMAKE_EXE_LINKER_FLAGS="--emit-symbol-map -sSTANDALONE_WASM=1 -sWASM=1 -sWASM_BIGINT=1 -sEXPORT_ALL=0 -sEXPORTED_RUNTIME_METHODS=cwrap,ccall,getValue,setValue,writeAsciiToMemory,wasmTable -lembind"
+  -DCMAKE_EXE_LINKER_FLAGS="--demangle --emit-symbol-map -sSTANDALONE_WASM=1 -sWASM=1 -sWASM_BIGINT=1 -sLINKABLE=1 -sEXPORT_ALL=1 -sEXPORTED_RUNTIME_METHODS=cwrap,ccall,getValue,setValue,writeAsciiToMemory,wasmTable --minify 0 -lembind"
   -DCMAKE_INSTALL_PREFIX="${LLVM_INSTALL_DIR}"
   -DCMAKE_SYSTEM_NAME=Emscripten
   -DCMAKE_TOOLCHAIN_FILE="$EMSDK/upstream/emscripten/cmake/Modules/Platform/Emscripten.cmake"
   -DCROSS_TOOLCHAIN_FLAGS_NATIVE="-DCMAKE_C_COMPILER=$CC;-DCMAKE_CXX_COMPILER=$CXX"
   -C "$TD/cmake/llvm_wasm_cache.cmake"
+  -DCMAKE_PROJECT_INCLUDE="$TD/cmake/llvm_wasm_project_include.cmake"
 )
 
 echo "Source Directory: ${LLVM_SOURCE_DIR}"
@@ -59,4 +61,10 @@ cmake --build "${LLVM_BUILD_DIR}" \
 sed -i.bak 's/CMAKE_EXECUTABLE_SUFFIX ".html"/CMAKE_EXECUTABLE_SUFFIX ".js"/g' "$EMSDK/upstream/emscripten/cmake/Modules/Platform/Emscripten.cmake"
 
 # wasi files aren't installed for some reason
-cp "${LLVM_BUILD_DIR}"/bin/* "${LLVM_INSTALL_DIR}/bin"
+cp "${LLVM_BUILD_DIR}"/bin/mlir-opt.* "${LLVM_INSTALL_DIR}/bin"
+# cp "${LLVM_BUILD_DIR}"/lib/*.symbols "${LLVM_INSTALL_DIR}/lib"
+
+# prevent symbol collisions
+# sed -i.bak 's/if(LLVM_ENABLE_PIC)/if(LLVM_ENABLE_PIC)\nreturn()\nendif()\nif(LLVM_ENABLE_PIC)/g' "${LLVM_SOURCE_DIR}/mlir/lib/ExecutionEngine/CMakeLists.txt"
+# https://stackoverflow.com/a/1252191/9045206
+# sed -i.bak -e ':a' -e 'N' -e '$!ba' -e 's/if(LLVM_ENABLE_PIC)\nreturn()\nendif()\n//g' "${LLVM_SOURCE_DIR}/mlir/lib/ExecutionEngine/CMakeLists.txt"

build_tools/cmake/llvm_wasm_cache.cmake

@@ -13,7 +13,15 @@ set(LLVM_ENABLE_LIBEDIT OFF CACHE BOOL "")
 set(LLVM_ENABLE_LIBPFM OFF CACHE BOOL "")
 set(LLVM_ENABLE_LIBXML2 OFF CACHE BOOL "")
 set(LLVM_ENABLE_OCAMLDOC OFF CACHE BOOL "")
+
+set(LLVM_BUILD_LLVM_DYLIB OFF CACHE BOOL "")
+set(MLIR_BUILD_MLIR_C_DYLIB OFF CACHE BOOL "")
+# when building libLLVM
+# relocation R_WASM_MEMORY_ADDR_SLEB cannot be used against symbol
 set(LLVM_ENABLE_PIC OFF CACHE BOOL "")
+set(MLIR_ENABLE_SPIRV_CPU_RUNNER OFF)
+set(MLIR_ENABLE_EXECUTION_ENGINE OFF)
+
 set(LLVM_ENABLE_THREADS OFF CACHE BOOL "")
 set(LLVM_ENABLE_UNWIND_TABLES OFF CACHE BOOL "")
 set(LLVM_ENABLE_ZLIB OFF CACHE BOOL "")
@@ -33,7 +41,7 @@ set(LLVM_MlirDevelopment_DISTRIBUTION_COMPONENTS
   llvm-headers
   llvm-libraries
   cmake-exports
-  opt
+  # opt
   # triggers LLVMMlirDevelopmentExports.cmake
   mlirdevelopment-cmake-exports
   # triggers MLIRMlirDevelopmentTargets.cmake
@@ -43,7 +51,7 @@ set(LLVM_MlirDevelopment_DISTRIBUTION_COMPONENTS
   mlir-headers
   mlir-libraries
   mlir-opt
-  mlir-reduce
-  mlir-tblgen
-  mlir-translate
+  # mlir-reduce
+  # mlir-tblgen
+  # mlir-translate
   CACHE STRING "")

build_tools/cmake/llvm_wasm_project_include.cmake

@@ -0,0 +1,27 @@
+# everything in here gets run at the end of the project loading
+# https://github.com/emscripten-core/emscripten/issues/15276#issuecomment-1039349267
+set_property(GLOBAL PROPERTY TARGET_SUPPORTS_SHARED_LIBS TRUE)
+# __ELF__ so that the correct LLVM_ABI macros get set
+# https://github.com/WebAssembly/tool-conventions/blob/main/BasicCABI.md#function-arguments-and-return-values
+# Similarly, types can either be returned directly from WebAssembly functions or returned indirectly via a pointer parameter prepended to the parameter list.
+# https://github.com/llvm/llvm-project/blob/6d973b4548e281d0b8e75e85833804bb45b6a0e8/clang/lib/CodeGen/Targets/WebAssembly.cpp#L135
+# https://github.com/llvm/llvm-project/commit/c285307e1457c4db2346443a4336e672d7487111#diff-b83bb889340990fea25762060e144b5cd4b4652a6fa737aaea9555d456344219
+# https://github.com/llvm/llvm-project/blame/63534779b4ef1816e2961246011e2ec3be110d27/clang/lib/CodeGen/TargetInfo.cpp
+set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D__ELF__")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__ELF__")
+set(CMAKE_SHARED_LIBRARY_CREATE_C_FLAGS "-sLINKABLE=1 -sEXPORT_ALL=1")
+set(CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS "-sLINKABLE=1 -sEXPORT_ALL=1")
+# prevent duplicated libs from being linked
+# will need to be renamed to DEDUPLICATION when CMake version catches up
+# https://github.com/Kitware/CMake/commit/5617c34c3135f7ec203d5a48b803eb323f458bc3#diff-17fc647759070cdaddd99e9ad994c4478860d9c301ed9a9a9f061a8825c8b690L20
+set(CMAKE_C_LINK_LIBRARIES_PROCESSING ORDER=FORWARD UNICITY=ALL)
+set(CMAKE_CXX_LINK_LIBRARIES_PROCESSING ORDER=FORWARD UNICITY=ALL)
+
+## hack to prevent -Bsymbolic
+#set(LLVM_LINKER_IS_SOLARISLD_ILLUMOS ON)
+
+set(CMAKE_SHARED_LIBRARY_SUFFIX ".wasm")
+set(CMAKE_STRIP FALSE)
+set(LLVM_NO_DEAD_STRIP ON)
+set(CMAKE_VERBOSE_MAKEFILE ON)
+set(CMAKE_CXX_VISIBILITY_PRESET default)

projects/js/embindings/CMakeLists.txt

@@ -0,0 +1,36 @@
+cmake_minimum_required(VERSION 3.29)
+project(eudsl-embindings LANGUAGES CXX C)
+set(CMAKE_C_STANDARD 11)
+set(CMAKE_CXX_STANDARD 17)
+
+if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR OR EUDSL_OUT_OF_TREE_BUILD)
+  find_package(MLIR REQUIRED CONFIG)
+  find_package(LLVM REQUIRED CONFIG)
+  message(STATUS "Using MLIRConfig.cmake in: ${MLIR_DIR}")
+  message(STATUS "Using LLVMConfig.cmake in: ${LLVM_DIR}")
+  set(LLVM_RUNTIME_OUTPUT_INTDIR ${CMAKE_BINARY_DIR}/bin)
+  set(LLVM_LIBRARY_OUTPUT_INTDIR ${CMAKE_BINARY_DIR}/lib)
+  set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
+  set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
+  list(APPEND CMAKE_MODULE_PATH "${MLIR_CMAKE_DIR}")
+  list(APPEND CMAKE_MODULE_PATH "${LLVM_CMAKE_DIR}")
+  include(TableGen)
+  include(AddLLVM)
+  include(AddMLIR)
+  include(HandleLLVMOptions)
+else()
+  set(MLIR_MAIN_SRC_DIR ${LLVM_MAIN_SRC_DIR}/../mlir)
+  set(MLIR_INCLUDE_DIR ${LLVM_MAIN_SRC_DIR}/../mlir/include)
+  set(MLIR_GENERATED_INCLUDE_DIR ${LLVM_BINARY_DIR}/tools/mlir/include)
+  set(MLIR_INCLUDE_DIRS "${MLIR_INCLUDE_DIR};${MLIR_GENERATED_INCLUDE_DIR}")
+endif()
+include_directories(${LLVM_INCLUDE_DIRS})
+include_directories(${MLIR_INCLUDE_DIRS})
+
+set_property(GLOBAL PROPERTY TARGET_SUPPORTS_SHARED_LIBS TRUE)
+set(CMAKE_SHARED_LIBRARY_CREATE_C_FLAGS "-s SIDE_MODULE=1")
+set(CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS "-s SIDE_MODULE=1")
+set(CMAKE_STRIP FALSE)  # used by default in pybind11 on .so modules
+
+add_library(example SHARED example.cpp)
+target_link_libraries(example PUBLIC MLIRIR)

projects/js/embindings/example.cpp

@@ -0,0 +1,37 @@
+//
+// Created by maksim on 11/14/24.
+//
+
+#include <emscripten/bind.h>
+#include <string>
+#include <vector>
+#include <optional>
+
+using namespace emscripten;
+
+std::vector<int> returnVectorData () {
+  std::vector<int> v(10, 1);
+  return v;
+}
+
+std::map<int, std::string> returnMapData () {
+  std::map<int, std::string> m;
+  m.insert(std::pair<int, std::string>(10, "This is a string."));
+  return m;
+}
+
+std::optional<std::string> returnOptionalData() {
+  return "hello";
+}
+
+EMSCRIPTEN_BINDINGS(module) {
+  function("returnVectorData", &returnVectorData);
+  function("returnMapData", &returnMapData);
+  function("returnOptionalData", &returnOptionalData);
+
+  // register bindings for std::vector<int>, std::map<int, std::string>, and
+  // std::optional<std::string>.
+  register_vector<int>("vector<int>");
+  register_map<int, std::string>("map<int, string>");
+  register_optional<std::string>();
+}