Fork bergamot-translator into this repostiory as inference

.clang-format

@@ -0,0 +1,5 @@
+BasedOnStyle: Google
+
+# Maximum line length 80 is too low even for 1080p monitor.  @XapaJIaMnu
+# personally would like 120.
+ColumnLimit: 120

.clang-format-ignore

@@ -0,0 +1,4 @@
+3rd_party
+wasm/test_page
+src/translator/aligned.h
+src/translator/pcqueue.h

.clang-tidy

@@ -0,0 +1,32 @@
+Checks: >
+  .*,
+  bugprone-*,  
+  concurrency-*,
+  google-*,
+  portability-*,
+  performance-*,
+  clang-analyzer-*,
+  readability-*,
+  -readability-implicit-bool-conversion,
+  -readability-isolate-declaration,
+  -readability-uppercase-literal-suffix,
+  misc-*,
+  -misc-noexcept*,
+  modernize-*,
+  -modernize-deprecated-headers,
+  -modernize-use-nodiscard,
+  -modernize-raw-string-literal,
+  -modernize-return-braced-init-list,
+  -modernize-use-equals-delete,
+  -modernize-use-trailing-return-type,
+
+
+
+CheckOptions:
+  - { key: readability-identifier-naming.ClassCase,     value: CamelCase  }
+  - { key: readability-identifier-naming.ClassMethodCase, value: camelBack  }
+  - { key: readability-identifier-naming.VariableCase,  value: camelBack  }
+  - { key: readability-identifier-naming.FunctionCase,  value: camelBack  }
+  - { key: readability-identifier-naming.PrivateMemberSuffix,  value: _   }
+  - { key: readability-identifier-naming.ParameterCase, value: camelBack  }
+

.github/CODEOWNERS

@@ -1,5 +1,31 @@
+# Firefox Translations review group
+.dockerignore @mozilla/firefox-translations
+.github @mozilla/firefox-translations
+.gitignore @mozilla/firefox-translations
+.gitmodules @mozilla/firefox-translations
+docker @mozilla/firefox-translations
+docs @mozilla/firefox-translations
+utils @mozilla/firefox-translations
+CODE_OF_CONDUCT.md @mozilla/firefox-translations
+LICENSE @mozilla/firefox-translations
+poetry.lock @mozilla/firefox-translations
+pyproject.toml @mozilla/firefox-translations
+README.md @mozilla/firefox-translations
+Taskfile.yml @mozilla/firefox-translations
+
+# Translations Training review group
+configs @mozilla/translations-training
+pipeline @mozilla/translations-training
+snakemake @mozilla/translations-training
+tests @mozilla/translations-training
+tracking @mozilla/translations-training
+
+# Translations Inference review group
+inference-engine @mozilla/translations-inference
+
 # Taskcluster pipeline related files. Changes to these ought to be reviewed by
 # RelEng to watch for security issues and best practices. These should also
 # be reviewed by people familiar with the pipeline itself.
-.taskcluster.yml @mozilla/releng
-taskcluster @mozilla/releng
+.taskcluster.yml @mozilla/releng @mozilla/translations-training
+taskcluster @mozilla/releng @mozilla/translations-training
+

.gitmodules

@@ -16,3 +16,27 @@
 [submodule "3rd_party/preprocess"]
 	path = 3rd_party/preprocess
 	url = https://github.com/kpu/preprocess.git
+[submodule "inference/3rd_party/ssplit-cpp"]
+	path = inference/3rd_party/ssplit-cpp
+	url = https://github.com/browsermt/ssplit-cpp
+# This is the same dependency and repository as `3rd_party/browsermt-marian-dev` below.
+#
+# When forking `inference-engine` into to this project, I made an earnest attempt to utilize the preexisting
+# `3rd_party/browsermt-marian-dev` submodule within `inference-engine`. Unfortunately, I ran into several roadblocks:
+#
+#   1) I cannot directly add `3rd_party/browsermt-marian-dev` as a cmake subdirectory because cmake is aware that
+#      this path is not a subdirectory of the `inference-engine` project root.
+#
+#   2) Symbolic links do not appear to work for git submodule direcotires the way that they do for regular directories.
+#      Even if the symbolic link had linked correctly, it may have still failed due to the considerations of 1).
+#
+#   3) I tried using cmake to copy the files from `3rd_party/browsermt-marian-dev` into `inference-engine/3rd_party/browsermt-marian-dev`
+#      at build time, which would ensure that there is no duplicate reference to the URL in this file, however the upstream dependency itself
+#      has hard-coded expectations that the `.git` directory is only one level up, which appears to work correctly for the way git submodules are
+#      configured, but does not work if the files are copied over to a regular directory deeper in the repository's directory tree.
+#
+# It may be possible to remove `3rd_party/browsermt-marian-dev` to instead use `inference-engine/3rd-party/browsermt-marian-dev` everywhere
+# within this repository, but I will leave that for a future commit if there is a need to do so.
+[submodule "inference/3rd_party/browsermt-marian-dev"]
+	path = inference/3rd_party/browsermt-marian-dev
+	url = https://github.com/browsermt/marian-dev

Taskfile.yml

@@ -75,6 +75,30 @@ tasks:
     cmds:
       - poetry run opuscleaner-server serve --host=0.0.0.0 --port=8000
 
+  inference-clean:
+    desc: Clean build artifacts from the inference directory.
+    cmds:
+      - >-
+          task docker-run -- ./inference/scripts/clean.sh
+
+  inference-build:
+    desc: Build inference engine.
+    cmds:
+      - >-
+          task docker-run -- ./inference/scripts/build-local.sh
+
+  inference-test:
+    desc: Run inference tests.
+    cmds:
+      - >-
+          task docker-run -- ./inference/scripts/unit-tests.sh
+
+  inference-build-wasm:
+    desc: Build inference engine WASM.
+    cmds:
+      - >-
+          task docker-run -- ./inference/scripts/build-wasm.sh
+
   lint-black:
     desc: Checks the styling of the Python code with Black.
     deps: [poetry-install-black]

inference/.gitignore

@@ -0,0 +1,30 @@
+# vim temporary files
+*.swp
+*.swo
+
+# CMake
+CMakeLists.txt.user
+CMakeCache.txt
+CMakeFiles
+CMakeScripts
+Testing
+Makefile
+cmake_install.cmake
+install_manifest.txt
+compile_commands.json
+CTestTestfile.cmake
+_deps
+
+
+wasm/test_page/node_modules
+/build
+/build-local
+/build-native
+/build-wasm
+/emsdk
+models
+wasm/module/worker/bergamot-translator-worker.*
+wasm/module/browsermt-bergamot-translator-*.tgz
+
+# VSCode
+.vscode

inference/3rd_party/CMakeLists.txt

@@ -0,0 +1,32 @@
+# browsermt-marian-dev is tested elsewhere in both paths, turning off here.
+set(COMPILE_TESTS OFF)
+add_subdirectory(browsermt-marian-dev EXCLUDE_FROM_ALL)
+
+if(COMPILE_WASM)
+  # This is a bad way of adding compilation flags. Will be improved soon.
+  add_compile_options(${WASM_COMPILE_FLAGS})
+  add_link_options(${WASM_LINK_FLAGS})
+endif(COMPILE_WASM)
+
+add_subdirectory(ssplit-cpp EXCLUDE_FROM_ALL)
+
+# Add include directories for 3rd party targets to be able to use it anywhere in the
+# project without explicitly specifying their include directories. Once they
+# fixe this problem, it can be removed.
+get_property(INCDIRS DIRECTORY browsermt-marian-dev/src PROPERTY INCLUDE_DIRECTORIES)
+target_include_directories(marian PUBLIC ${INCDIRS})
+
+get_property(INCLUDE_DIRECTORIES DIRECTORY ssplit-cpp/src PROPERTY INCLUDE_DIRECTORIES)
+target_include_directories(ssplit PUBLIC ${INCLUDE_DIRECTORIES})
+
+get_property(COMPILE_DEFINITIONS DIRECTORY browsermt-marian-dev PROPERTY COMPILE_DEFINITIONS) 
+target_compile_definitions(marian PUBLIC ${COMPILE_DEFINITIONS})
+
+get_property(COMPILE_OPTIONS DIRECTORY browsermt-marian-dev PROPERTY COMPILE_OPTIONS) 
+target_compile_options(marian PUBLIC ${COMPILE_OPTIONS})
+
+# Compilation flags 
+get_directory_property(CMAKE_C_FLAGS DIRECTORY browsermt-marian-dev DEFINITION CMAKE_C_FLAGS) 
+get_directory_property(CMAKE_CXX_FLAGS DIRECTORY browsermt-marian-dev DEFINITION CMAKE_CXX_FLAGS) 
+set(CMAKE_C_FLAGS ${CMAKE_C_FLAGS} PARENT_SCOPE)    
+set(CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS} PARENT_SCOPE)

inference/BERGAMOT_VERSION

@@ -0,0 +1 @@
+v0.4.5

inference/CMakeLists.txt

@@ -0,0 +1,188 @@
+cmake_minimum_required(VERSION 3.5.1)
+set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
+
+if (POLICY CMP0074)
+  cmake_policy(SET CMP0074 NEW) # CMake 3.12
+endif ()
+
+if (POLICY CMP0077)
+  cmake_policy(SET CMP0077 NEW)
+endif()
+
+project(bergamot_translator CXX C)
+
+# Retrieve the parent-directory path of PROJECT_SOURCE_DIR and assign that to REPOSITORY_ROOT_DIR.
+cmake_path(GET PROJECT_SOURCE_DIR PARENT_PATH REPOSITORY_ROOT_DIR)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+# Generate a compile_commands.json in the build directory. The compile commands allow
+# code editors to understand the build process and provide static analysis of the code.
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+# Note that with CMake MSVC build, the option CMAKE_BUILD_TYPE is automatically derived from the key
+# 'configurationType' in CMakeSettings.json configurations
+if(NOT CMAKE_BUILD_TYPE)
+  message(WARNING "CMAKE_BUILD_TYPE not set; setting to Release")
+  set(CMAKE_BUILD_TYPE "Release")
+endif()
+
+if(NOT COMPILE_WASM)
+  # Setting BUILD_ARCH to native invokes CPU intrinsic detection logic below.
+  # Prevent invoking that logic for WASM builds.
+  set(BUILD_ARCH native CACHE STRING "Compile for this CPU architecture.")
+
+  # Unfortunately MSVC supports a limited subset of BUILD_ARCH flags. Instead try to guess
+  # what architecture we can compile to reading BUILD_ARCH and mapping it to MSVC values
+  # references: https://clang.llvm.org/docs/UsersManual.html https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html https://gcc.gnu.org/onlinedocs/gcc-4.8.5/gcc/i386-and-x86-64-Options.html
+  # https://docs.microsoft.com/en-us/cpp/build/reference/arch-x86?redirectedfrom=MSDN&view=vs-2019&view=msvc-170 https://devblogs.microsoft.com/oldnewthing/20201026-00/?p=104397
+  # This is by no means an exhaustive list but should match the most common flags Linux programmers expect to parse to MSVC
+  if(MSVC)
+    if(BUILD_ARCH STREQUAL "native") # avx2 is good default for native. Very few desktop systems support avx512
+      set(MSVC_BUILD_ARCH "/arch:AVX2")
+    elseif(BUILD_ARCH STREQUAL "skylake-avx512" OR BUILD_ARCH STREQUAL "cannonlake" OR BUILD_ARCH STREQUAL "x86-64-v4" OR BUILD_ARCH STREQUAL "tigerlake" OR BUILD_ARCH STREQUAL "cooperlake" OR BUILD_ARCH STREQUAL "cascadelake")
+      set(MSVC_BUILD_ARCH "/arch:AVX512")
+    elseif(BUILD_ARCH STREQUAL "core-avx2" OR BUILD_ARCH STREQUAL "haswell" OR BUILD_ARCH STREQUAL "x86-64-v3" OR BUILD_ARCH STREQUAL "broadwell" OR BUILD_ARCH STREQUAL "skylake")
+      set(MSVC_BUILD_ARCH "/arch:AVX2")
+    elseif(BUILD_ARCH STREQUAL "sandybridge" OR BUILD_ARCH STREQUAL "corei7-avx" OR BUILD_ARCH STREQUAL "core-avx-i" OR BUILD_ARCH STREQUAL "ivybridge")
+      set(MSVC_BUILD_ARCH "/arch:AVX")
+    elseif(BUILD_ARCH STREQUAL "nehalem" OR BUILD_ARCH STREQUAL "westmere" OR BUILD_ARCH STREQUAL "x86-64-v2" OR BUILD_ARCH STREQUAL "corei7" OR BUILD_ARCH STREQUAL "core2")
+      set(MSVC_BUILD_ARCH "/arch:SSE2") # This is MSVC default. We won't go down to SSE because we don't support that hardware at all with intgemm. Marian recommends to only go down to SSE4.1 at most
+    else()
+      message(WARNING "Unknown BUILD_ARCH ${BUILD_ARCH} provided. Default to SSE2 for Windows build")
+      set(MSVC_BUILD_ARCH "/arch:SSE2")
+    endif()
+  endif(MSVC)
+endif()
+
+#MSVC can't seem to pick up correct flags otherwise:
+if(MSVC)
+  add_definitions(-DUSE_SSE2=1) # Supposed to fix something in the sse_mathfun.h but not sure it does
+  set(INTRINSICS ${MSVC_BUILD_ARCH}) # ARCH we're targetting on win32. @TODO variable
+
+  set(CMAKE_CXX_FLAGS           "/EHsc /DWIN32 /D_WINDOWS /DUNICODE /D_UNICODE /D_CRT_NONSTDC_NO_WARNINGS /D_CRT_SECURE_NO_WARNINGS /bigobj")
+  set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS} /MT /O2 ${INTRINSICS} /MP /GL /DNDEBUG")
+  set(CMAKE_CXX_FLAGS_DEBUG     "${CMAKE_CXX_FLAGS} /MTd /Od /Ob0 ${INTRINSICS} /RTC1 /Zi /D_DEBUG")
+
+  # ignores warning LNK4049: locally defined symbol free imported - this comes from zlib
+  set(CMAKE_EXE_LINKER_FLAGS         "${CMAKE_EXE_LINKER_FLAGS} /DEBUG /LTCG:incremental /INCREMENTAL:NO /ignore:4049")
+  set(CMAKE_EXE_LINKER_FLAGS_RELEASE "${CMAKE_EXE_LINKER_FLAGS} /NODEFAULTLIB:MSVCRT")
+  set(CMAKE_EXE_LINKER_FLAGS_DEBUG   "${CMAKE_EXE_LINKER_FLAGS} /NODEFAULTLIB:MSVCRTD")
+  set(CMAKE_STATIC_LINKER_FLAGS      "${CMAKE_STATIC_LINKER_FLAGS} /LTCG:incremental")
+endif(MSVC)
+
+include(CMakeDependentOption)
+
+# Project specific cmake options
+option(COMPILE_WASM "Compile for WASM" OFF)
+cmake_dependent_option(USE_WASM_COMPATIBLE_SOURCE "Use wasm compatible sources" OFF "NOT COMPILE_WASM" ON)
+
+# WASM disables a million libraries, which also includes the unit test-library.
+cmake_dependent_option(COMPILE_UNIT_TESTS "Compile unit tests" OFF "USE_WASM_COMPATIBLE_SOURCE" ON)
+option(COMPILE_TESTS "Compile bergamot-tests" OFF)
+cmake_dependent_option(ENABLE_CACHE_STATS "Enable stats on cache" ON "COMPILE_TESTS" OFF)
+
+
+# Set 3rd party submodule specific cmake options for this project
+SET(COMPILE_CUDA OFF CACHE BOOL "Compile GPU version")
+SET(USE_SENTENCEPIECE ON CACHE BOOL "Download and compile SentencePiece")
+SET(USE_STATIC_LIBS ON CACHE BOOL "Link statically against non-system libs")
+SET(SSPLIT_COMPILE_LIBRARY_ONLY ON CACHE BOOL "Do not compile ssplit tests")
+if (USE_WASM_COMPATIBLE_SOURCE)
+  SET(COMPILE_LIBRARY_ONLY ON CACHE BOOL "Build only the Marian library and exclude all executables.")
+  SET(USE_MKL OFF CACHE BOOL "Compile with MKL support")
+  # # Setting the ssplit-cpp submodule specific cmake options for wasm
+  SET(SSPLIT_USE_INTERNAL_PCRE2 ON CACHE BOOL "Use internal PCRE2 instead of system PCRE2")
+endif()
+
+# Documentation: https://cliutils.gitlab.io/modern-cmake/chapters/projects/submodule.html
+# Ensures the submodules are set correctly during a build.
+find_package(Git QUIET)
+if(GIT_FOUND AND EXISTS "${REPOSITORY_ROOT_DIR}/.git")
+# Update submodules as needed
+    option(GIT_SUBMODULE "Check submodules during build" ON)
+    if(GIT_SUBMODULE)
+        message(STATUS "Submodule update")
+        execute_process(COMMAND ${GIT_EXECUTABLE} submodule update --init --recursive
+                        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
+                        RESULT_VARIABLE GIT_SUBMOD_RESULT)
+        if(NOT GIT_SUBMOD_RESULT EQUAL "0")
+            message(FATAL_ERROR "git submodule update --init failed with ${GIT_SUBMOD_RESULT}, please checkout submodules")
+        endif()
+    endif()
+endif()
+
+# Project versioning
+include(GetVersionFromFile)
+message(STATUS "Project name: ${PROJECT_NAME}")
+message(STATUS "Project version: ${PROJECT_VERSION_STRING_FULL}")
+
+if(COMPILE_WASM)
+  # See https://github.com/emscripten-core/emscripten/blob/main/src/settings.js
+  list(APPEND WASM_COMPILE_FLAGS
+    -O3
+    # Preserve whitespaces in JS even for release builds; this doesn't increase wasm binary size
+    $<$<CONFIG:Release>:-g1>
+    # Relevant Debug info only for release with debug builds as this increases wasm binary size
+    $<$<CONFIG:RelWithDebInfo>:-g2>
+    -fPIC
+    -mssse3
+    -msimd128
+    # -fno-exceptions # Can't do that because spdlog uses exceptions
+    -sDISABLE_EXCEPTION_CATCHING=1
+    -sSTRICT=1
+  )
+  list(APPEND WASM_LINK_FLAGS
+    -O3
+    # Preserve whitespaces in JS even for release builds; this doesn't increase wasm binary size
+    $<$<CONFIG:Release>:-g1>
+    # Relevant Debug info only for release with debug builds as this increases wasm binary size
+    $<$<CONFIG:RelWithDebInfo>:-g2>
+    -lembind
+    # Save some code, and some speed
+    -sASSERTIONS=0
+    -sDISABLE_EXCEPTION_CATCHING=1
+    # the intgemm functions we call will be undefined since these are linked at
+    # runtime by our own javascript.
+    -sLLD_REPORT_UNDEFINED
+    -sERROR_ON_UNDEFINED_SYMBOLS=0
+    # Cause we can!
+    -sSTRICT=1
+    # You know we need it
+    -sALLOW_MEMORY_GROWTH=1
+    -sENVIRONMENT=web,worker
+    # No need to call main(), there's nothing there.
+    -sINVOKE_RUN=0
+    # No need for filesystem code in the generated Javascript
+    -sFILESYSTEM=0
+    # If you turn this on, it will mangle names which makes the dynamic linking hard.
+    -sDECLARE_ASM_MODULE_EXPORTS=0
+    # Export all of the intgemm functions in case we need to fall back to using the embedded intgemm
+    -sEXPORTED_FUNCTIONS=[_int8PrepareAFallback,_int8PrepareBFallback,_int8PrepareBFromTransposedFallback,_int8PrepareBFromQuantizedTransposedFallback,_int8PrepareBiasFallback,_int8MultiplyAndAddBiasFallback,_int8SelectColumnsOfBFallback]
+    # Necessary for mozintgemm linking. This prepares the `wasmMemory` variable ahead of time as
+    # opposed to delegating that task to the wasm binary itself. This way we can link MozIntGEMM
+    # module to the same memory as the main bergamot-translator module.
+    -sIMPORTED_MEMORY=1
+    # Dynamic execution is either frowned upon or blocked inside browser extensions
+    -sDYNAMIC_EXECUTION=0
+  )
+endif(COMPILE_WASM)
+
+# Needs to be enabled before including the folder containing tests (src/tests)
+if(COMPILE_TESTS)
+    enable_testing()
+endif(COMPILE_TESTS)
+
+add_subdirectory(3rd_party)
+add_subdirectory(src)
+
+if(COMPILE_WASM)
+  add_subdirectory(wasm)
+endif(COMPILE_WASM)
+
+option(COMPILE_PYTHON "Compile python bindings. Intended to be activated with setup.py" OFF)
+if(COMPILE_PYTHON)
+  add_subdirectory(bindings/python)
+endif(COMPILE_PYTHON)
+