Change defaults and improve CMake export

jerinphilip · Oct 28, 2023 · 4322e41 · 4322e41
1 parent ed4bfa4
commit 4322e41
Show file tree

Hide file tree

Showing 16 changed files with 226 additions and 104 deletions.
diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
@@ -32,6 +32,7 @@ jobs:
           run: |
             sudo apt-get update
             sudo apt-get install -y build-essential cmake
+
             python3 -m pip install black isort
             python3 -m pip install cmake-format
             sudo apt-get install -y shfmt
@@ -71,7 +72,7 @@ jobs:
           with:
             submodules: recursive
 
-        - name: Install dependencies
+        - name: Setup
           run: |
             bash scripts/ci/${{ matrix.name }}/01-setup.sh
 
@@ -87,6 +88,15 @@ jobs:
           run:
             bash scripts/ci/${{ matrix.name }}/04-package.sh
 
+        - uses: actions/upload-artifact@v2
+          if: ${{ failure() }}
+          with:
+            name: slimt
+            path: |
+              ${{ github.workspace }}/slimt-coredump
+
+
+
   build-wheels:
       strategy:
         matrix:
@@ -199,7 +209,7 @@ jobs:
               PYTHON_LOCAL_VERSION_IDENTIFIER=${{ env.PYTHON_LOCAL_VERSION_IDENTIFIER }}
               CCACHE_DIR=/host/${{ env.ccache_dir }}
               CCACHE_BASEDIR=/host/${{ env.ccache_basedir }}
-              CMAKE_ARGS="-DUSE_PYBIND11_SOURCE=ON -DBLA_VENDOR=OpenBLAS"
+              CMAKE_ARGS="-DUSE_PYBIND11_SOURCE=ON -DBLA_VENDOR=OpenBLAS -DUSE_BUILTIN_SENTENCEPIECE=ON -DWITH_GEMMOLOGY=OFF -DWITH_INTGEMM=ON -DUSE_SSE2=ON"
 
             CIBW_ENVIRONMENT_MACOS:
               USE_CCACHE=1
@@ -232,9 +242,11 @@ jobs:
               
               # Install CBLAS
               # yum install -y atlas-devel blas-devel
-
+              
               chmod -R a+rwx /host/${{ env.ccache_dir }}
 
+              lscpu
+
               ccache -s # Print current cache stats
               ccache -z # Zero cache entry
 

diff --git a/3rd-party/CMakeLists.txt b/3rd-party/CMakeLists.txt
@@ -31,10 +31,9 @@ else(USE_BUILTIN_SENTENCEPIECE)
   # surface, it becomes a problem. See:
   # https://github.com/google/sentencepiece/issues/922
   add_library(sentencepiece_pb INTERFACE)
-  target_include_directories(sentencepiece_pb
-                             INTERFACE sentencepiece/third_party/protobuf-lite)
-  target_include_directories(sentencepiece_pb
-                             INTERFACE sentencepiece/src/builtin_pb)
+  target_include_directories(
+    sentencepiece_pb INTERFACE sentencepiece/third_party/protobuf-lite
+                               sentencepiece/src/builtin_pb)
 
   add_library(SentencePiece::SentencePiece ALIAS sentencepiece)
   add_library(SentencePiece::Protobuf ALIAS sentencepiece_pb)

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -13,9 +13,9 @@ set(CMAKE_VISIBILITY_INLINE_HIDDEN YES)
 option(WITH_ASAN "Enable AddressSanitizer in builds" OFF)
 option(WITH_TESTS "Enable building tests (requires a few extra libraries)" OFF)
 
-option(WITH_INTGEMM "Use intgemm" ON)
+option(WITH_INTGEMM "Use intgemm" OFF)
 option(WITH_RUY "Use ruy" OFF)
-option(WITH_GEMMOLOGY "Use gemmology" OFF)
+option(WITH_GEMMOLOGY "Use gemmology" ON)
 option(WITH_BLAS "Use BLAS. Otherwise moves to ruy" ON)
 
 option(USE_BUILTIN_SENTENCEPIECE "Use SentencePiece supplied as 3rd-party" ON)
@@ -26,7 +26,7 @@ option(USE_SSSE3 "Use SSSE3" OFF)
 option(USE_SSE2 "Use SSE2" OFF)
 option(USE_NEON "Use NEON" OFF)
 
-option(SLIMT_SIMD "Use SLIMT SIMD" OFF)
+option(SLIMT_PACKAGE "Package for cmake, pkgconfig" OFF)
 
 set(SLIMT_COMPILE_OPTIONS
     -Wall
@@ -51,30 +51,41 @@ if(WITH_ASAN)
   list(APPEND SLIMT_LINK_OPTIONS -fsanitize=address -fsanitize=undefined)
 endif(WITH_ASAN)
 
-find_package(PCRE2 REQUIRED)
+if(WITH_TCMALLOC)
+  find_library(
+    TCMALLOC_LIBRARIES
+    NAMES tcmalloc_minimal
+    PATHS /usr/lib /usr/local/lib)
 
-if(USE_BUILTIN_SENTENCEPIECE)
-  list(APPEND SLIMT_PUBLIC_LIBS SentencePiece::SentencePiece)
-else(USE_BUILTIN_SENTENCEPIECE)
-  list(APPEND SLIMT_PRIVATE_LIBS SentencePiece::SentencePiece)
-  list(APPEND SLIMT_PRIVATE_LIBS SentencePiece::Protobuf)
-endif(USE_BUILTIN_SENTENCEPIECE)
+  if(TCMALLOC_LIBRARIES)
+    message(STATUS "tcmalloc found: ${TCMALLOC_LIBRARIES}")
+  else(TCMALLOC_LIBRARIES)
+    message(FATAL_ERROR "tcmalloc not found")
+  endif(TCMALLOC_LIBRARIES)
 
-list(APPEND SLIMT_PRIVATE_LIBS PCRE2::PCRE2)
+  list(APPEND SLIMT_PRIVATE_LIBS ${TCMALLOC_LIBRARIES})
+endif(WITH_TCMALLOC)
 
+find_package(PCRE2 REQUIRED)
 add_subdirectory(3rd-party)
 
+list(APPEND SLIMT_PRIVATE_LIBS PCRE2::PCRE2)
+list(APPEND SLIMT_PUBLIC_LIBS SentencePiece::SentencePiece)
+
+if(NOT USE_BUILTIN_SENTENCEPIECE)
+  list(APPEND SLIMT_PRIVATE_LIBS SentencePiece::Protobuf)
+endif(NOT USE_BUILTIN_SENTENCEPIECE)
+
 if(WITH_BLAS)
   find_package(BLAS REQUIRED)
-  find_path(CBLAS_INCLUDE_DIR cblas.h)
 
+  # The above defines BLAS::BLAS only one recent cmakes, the following code
+  # detects if the target is available if not creates an alias target.
   if(BLAS_FOUND AND NOT TARGET BLAS::BLAS)
     add_library(BLAS::BLAS INTERFACE IMPORTED)
     set_target_properties(
-      BLAS::BLAS
-      PROPERTIES INTERFACE_INCLUDE_DIRECTORIES
-                 "${BLAS_INCLUDE_DIR} ${CBLAS_INCLUDE_DIR}"
-                 INTERFACE_LINK_LIBRARIES "${BLAS_LIBRARIES}")
+      BLAS::BLAS PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${BLAS_INCLUDE_DIR} "
+                            INTERFACE_LINK_LIBRARIES "${BLAS_LIBRARIES}")
   endif()
 
   list(APPEND SLIMT_PRIVATE_LIBS BLAS::BLAS)
@@ -164,26 +175,24 @@ message(STATUS "Project version: ${PROJECT_VERSION_STRING_FULL}")
 
 if(UNIX)
   include(GNUInstallDirs)
-  if(BUILD_CPACK)
-    # cmake-format: off
-    set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
-    set(CPACK_PACKAGE_CONTACT "Jerin Philip <[email protected]>")
-    set(CPACK_DEBIAN_PACKAGE_MAINTAINER "Jerin Philip")
-    set(CPACK_PACKAGE_DIRECTORY ${CMAKE_BINARY_DIR}/deb)
-    set(CPACK_GENERATOR "DEB")
-    set(CPACK_DEBIAN_PACKAGE_DEPENDS libsentencepiece-dev)
-    set(CPACK_PACKAGE_NAME ${PROJECT_NAME})
-    set(CPACK_PACKAGE_FILE_NAME ${PROJECT_NAME})
-    set(CPACK_PACKAGE_VERSION ${PROJECT_VERSION_STRING})
-    set(CPACK_DEBIAN_PACKAGE_NAME ${PROJECT_NAME})
-    set(CPACK_DEBIAN_PACKAGE_VERSION "${PROJECT_VERSION_STRING}")
-    set(CPACK_PACKAGE_VERSION_MAJOR ${PROJECT_VERSION_MAJOR})
-    set(CPACK_PACKAGE_VERSION_MINOR ${PROJECT_VERSION_MINOR})
-    set(CPACK_PACKAGE_VERSION_PATCH ${PROJECT_VERSION_PATCH})
-    set(CPACK_DEBIAN_PACKAGE_DESCRIPTION "slimt: Inference library for tiny translation models.")
-    # cmake-format: on
-    include(CPack)
-  endif(BUILD_CPACK)
+  # cmake-format: off
+  set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
+  set(CPACK_PACKAGE_CONTACT "Jerin Philip <[email protected]>")
+  set(CPACK_DEBIAN_PACKAGE_MAINTAINER "Jerin Philip")
+  set(CPACK_PACKAGE_DIRECTORY ${CMAKE_BINARY_DIR}/deb)
+  set(CPACK_GENERATOR "DEB")
+  set(CPACK_DEBIAN_PACKAGE_DEPENDS libsentencepiece-dev)
+  set(CPACK_PACKAGE_NAME ${PROJECT_NAME})
+  set(CPACK_PACKAGE_FILE_NAME ${PROJECT_NAME})
+  set(CPACK_PACKAGE_VERSION ${PROJECT_VERSION_STRING})
+  set(CPACK_DEBIAN_PACKAGE_NAME ${PROJECT_NAME})
+  set(CPACK_DEBIAN_PACKAGE_VERSION "${PROJECT_VERSION_STRING}")
+  set(CPACK_PACKAGE_VERSION_MAJOR ${PROJECT_VERSION_MAJOR})
+  set(CPACK_PACKAGE_VERSION_MINOR ${PROJECT_VERSION_MINOR})
+  set(CPACK_PACKAGE_VERSION_PATCH ${PROJECT_VERSION_PATCH})
+  set(CPACK_DEBIAN_PACKAGE_DESCRIPTION "slimt: Inference library for tiny translation models.")
+  # cmake-format: on
+  include(CPack)
 endif(UNIX)
 
 add_subdirectory(slimt)

diff --git a/README.md b/README.md
@@ -49,19 +49,24 @@ More information on the models are described in the following papers:
 The large-list of dependencies from bergamot-translator have currently been
 reduced to:
 
-* For `int8_t` matrix-multiply [intgemm](https://github.com/kpu/intgemm) (`x86_64`) or
-  [ruy](https://github.com/google/ruy) (`aarch64`).
+* For `int8_t` matrix-multiply [intgemm](https://github.com/kpu/intgemm)
+  (`x86_64`) or [ruy](https://github.com/google/ruy) (`aarch64`) or
+  [xsimd](https://github.com/xtensor-stack/xsimd) via
+  [gemmology](https://github.com/mozilla/gemmology).
 * For vocabulary - [sentencepiece](https://github.com/browsermt/sentencepiece). 
-* For `sgemm` - Whatever BLAS provider is found via CMake. 
-* CLI11 (only a dependency for cmdline) 
+* For sentence-splitting using regular-expressions
+  [PCRE2](https://github.com/PCRE2Project/pcre2).
+* For `sgemm` - Whatever BLAS provider is found via CMake (openblas,
+  intel-oneapimkl, cblas).  Feel free to provide
+  [hints](https://cmake.org/cmake/help/latest/module/FindBLAS.html#blas-lapack-vendors). 
+* [CLI11](https://github.com/CLIUtils/CLI11/) (only a dependency for cmdline) 
 
 Source code is made public where basic functionality (text-translation) works
 for English-German tiny models. Parity in features and speed with marian and
 bergamot-translator (where relevant) is a work-in-progress. Eventual support for
 `base` models are planned. Contributions are welcome and appreciated.
 
 
-
 ## Getting started
 
 Clone with submodules.
@@ -70,32 +75,86 @@ Clone with submodules.
 git clone --recursive https://github.com/jerinphilip/slimt.git
 ```
 
-Configure and build.
+Configure and build. `slimt` is still experimenting with CMake and
+dependencies. The following should work at the moment:
+
 
 ```bash
-# Configure
-cmake -B build -S $PWD -DCMAKE_BUILD_TYPE=Release 
+
+# Configure intgemm
+cmake -B build -S $PWD -DCMAKE_BUILD_TYPE=Release -DWITH_INTGEMM=ON 
+# Configure ruy instead of intgemm
+cmake -B build -S $PWD -DCMAKE_BUILD_TYPE=Release -DWITH_RUY=ON
 
 # Build
 cmake --build build --target all --parallel 4
 ```
 
-Successful build generate two executables `slimt` and `slimt_test` for
+Successful build generate two executables `slimt-cli` and `slimt-test` for
 command-line usage and testing respectively.
 
 ```bash
-build/bin/slimt                               \
+build/bin/slimt-cli                           \
     --root <path/to/folder>                   \
     --model </relative/path/to/model>         \
     --vocabulary </relative/path/to/vocab>    \
     --shortlist </relative/path/to/shortlist>
 
-build/slimt_test <test-name>
+build/slimt-test <test-name>
+```
+
+### Distribution
+
+There is a build-path being prepared towards packaging on Linux. To use this,
+configure with the following args:
+
+```bash
+# Configure to use xsimd via gemmology
+ARGS=(
+    # Use gemmology
+    -DWITH_GEMMOLOGY=ON               
+
+    # -DUSE_AVX512 -DUSE_SSSE3 ... -DUSE_NEON also available.
+    -DUSE_AVX2=ON                          
+
+    # Use sentencepiece installed via system.
+    -DUSE_BUILTIN_SENTENCEPIECE=OFF        
+
+    # Exports slimtConfig.cmake (cmake) and slimt.pc.in (pkg-config)
+    -DSLIMT_PACKAGE=ON 
+
+    -DCMAKE_INSTALL_PREFIX=/path/to/prefix
+)
+
+cmake -B build -S $PWD -DCMAKE_BUILD_TYPE=Release "${ARGS[@]}"
+cmake --build build --target all
+
+# May require if prefix is writable only by root.
+cmake --build build --target install 
+```
+
+The above run expects the packages `sentencepiece`, `xsimd` and a BLAS provider
+to come from the system's package manager. Examples of this in distributions
+include:
+
+```bash
+# Debian based systems
+sudo apt-get install -y libxsimd-dev libsentencepiece-dev libopenblas-dev
+
+# ArchLinux
+pacman -S openblas xsimd
+yay -S sentencepiece-git
 ```
 
+This is still very much a work in progress, towards being able to make
+[lemonade](https://github.com/jerinphilip/lemonade) available in distributions.
+Help is much appreciated here, please get in touch if you can help here.
+
 ### Python
 
-Python bindings to the C++ code are available. 
+Python bindings to the C++ code are available.  Python bindings provide a layer
+to download models and use-them via command line entrypoint `slimt` (the core
+slimt library only has the inference code).
 
 ```bash
 python3 -m venv env
@@ -108,3 +167,9 @@ python3 -m pip install dist/<wheel-name>.whl
 slimt download -m en-de-tiny
 slimt download -m de-en-tiny
 ```
+
+You may pass customizing cmake-variables via `CMAKE_ARGS` environment variable.
+
+```bash
+CMAKE_ARGS='-D...' python3 setup.py bdist_wheel
+```
diff --git a/cmake/slimtConfig.cmake.in b/cmake/slimtConfig.cmake.in
@@ -7,11 +7,3 @@ pkg_check_modules(Sentencepiece REQUIRED sentencepiece)
 
 # Exports slimt::slimt-shared target.
 include("${CMAKE_CURRENT_LIST_DIR}/slimtTargets.cmake")
-
-# Sentencepiece has to be picked up by CMake, as it is not exported by
-# slimtTargets.cmake.
-add_library(slimt_export INTERFACE)
-target_link_libraries(slimt_export INTERFACE slimt::slimt-shared
-                                             ${Sentencepiece_LIBRARIES})
-target_include_directories(slimt_export INTERFACE ${Sentencepiece_INCLUDE_DIRS})
-add_library(slimt::slimt ALIAS slimt_export)
diff --git a/scripts/ci/android/02-build.sh b/scripts/ci/android/02-build.sh
@@ -14,6 +14,8 @@ function cmake-configure {
   SLIMT_ARGS=(
     -DWITH_RUY=ON
     -DWITH_INTGEMM=OFF
+    -DWITH_GEMMOLOGY=OFF
+    -DUSE_BUILTIN_SENTENCEPIECE=ON
     -DWITH_BLAS=OFF
     -DSLIMT_USE_INTERNAL_PCRE2=ON
   )

diff --git a/scripts/ci/format-check.sh b/scripts/ci/format-check.sh
@@ -19,7 +19,6 @@ function slimt-check-clang-tidy {
 
     # Gemmology, which is default on has to be turned off.
     -DWITH_GEMMOLOGY=OFF
-    -DEXPORT_CMAKE_FILE=OFF
   )
 
   cmake -B build -S . "${ARGS[@]}"

diff --git a/scripts/ci/macos/01-setup.sh b/scripts/ci/macos/01-setup.sh
@@ -1,3 +1,16 @@
 #!/bin/bash
 
 brew install cmake
+brew install xsimd openblas
+brew install sentencepiece
+
+sysctl -a | grep machdep.cpu.features
+
+ulimit -c unlimited # Enable core dumps to be captured (must be in same run block)
+
+COREDUMP_DIR="$PWD/slimt-coredump"
+mkdir -p "${COREDUMP_DIR}"
+COREDUMP_PATTERN="${COREDUMP_DIR}/core.%n.%P.%t"
+sudo sysctl -w kern.corefile=${COREDUMP_PATTERN}
+
+echo "coredumps: ${COREDUMP_PATTERN}"
diff --git a/scripts/ci/macos/02-build.sh b/scripts/ci/macos/02-build.sh
@@ -3,7 +3,18 @@
 set -eo pipefail
 
 # Configure
-cmake -B build -S $PWD -DCMAKE_BUILD_TYPE=Release -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DWITH_INTGEMM=ON -DWITH_RUY=OFF
+ARGS=(
+  -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
+  -DWITH_INTGEMM=OFF -DWITH_RUY=OFF
+  -DWITH_GEMMOLOGY=ON
+  -DUSE_SSE2=ON
+  -DUSE_BUILTIN_SENTENCEPIECE=OFF
+
+  -DCMAKE_BUILD_TYPE=Debug
+  -DWITH_ASAN=ON
+)
+
+cmake -B build -S $PWD "${ARGS[@]}"
 
 # Build
 cmake --build build --target all