add cpu mode for built-in Faiss (#841)

* add cpu mode for built-in Faiss

* delete customization options on build stage

* delete faiss version

* delete customization options on build stage

* fix unittest problem in cpu mode

* fix unittest problem in cpu mode

Co-authored-by: quicksilver <zhifeng.zhang@zilliz.com>
Co-authored-by: Jin Hai <hai.jin@zilliz.com>

CHANGELOG.md

@@ -22,6 +22,7 @@ Please mark all change in change log and use the issue from GitHub
 - \#738 - Use Openblas / lapack from apt install
 - \#758 - Enhance config description
 - \#791 - Remove Arrow
+- \#834 - add cpu mode for built-in Faiss
 - \#848 - Add ready-to-use config files to the Milvus repo for enhanced user experince
 
 ## Task

ci/jenkins/step/build.groovy

@@ -4,7 +4,7 @@ timeout(time: 75, unit: 'MINUTES') {
             def checkResult = sh(script: "./check_ccache.sh -l ${params.JFROG_ARTFACTORY_URL}/ccache", returnStatus: true)
 
             if ("${BINARY_VERSION}" == "gpu") {
-                sh "/bin/bash --login -c \". ./before-install.sh && ./build.sh -t ${params.BUILD_TYPE} -o ${env.MILVUS_INSTALL_PREFIX} -l -g -x -u -c\""
+                sh "/bin/bash --login -c \". ./before-install.sh && ./build.sh -t ${params.BUILD_TYPE} -o ${env.MILVUS_INSTALL_PREFIX} -l -g -u -c\""
             } else {
                 sh "/bin/bash --login -c \". ./before-install.sh && ./build.sh -t ${params.BUILD_TYPE} -o ${env.MILVUS_INSTALL_PREFIX} -l -u -c\""
             }

ci/scripts/build.sh

@@ -17,14 +17,13 @@ BUILD_UNITTEST="OFF"
 INSTALL_PREFIX="/var/lib/milvus"
 FAISS_ROOT=""
 PRIVILEGES="OFF"
-CUSTOMIZATION="OFF" # default use origin faiss
 BUILD_COVERAGE="OFF"
 RUN_CPPLINT="OFF"
 GPU_VERSION="OFF"
 WITH_MKL="OFF"
 CUDA_COMPILER=/usr/local/cuda/bin/nvcc
 
-while getopts "o:t:b:f:pgxulcmh" arg
+while getopts "o:t:b:f:pgulcmh" arg
 do
         case $arg in
              o)
@@ -49,9 +48,6 @@ do
                 echo "Build and run unittest cases" ;
                 BUILD_UNITTEST="ON";
                 ;;
-             x)
-                CUSTOMIZATION="ON";
-                ;;
              l)
                 RUN_CPPLINT="ON"
                 ;;
@@ -71,7 +67,6 @@ parameter:
 -f: faiss root path
 -p: install command with elevated privileges
 -g: gpu version
--x: milvus customization (default: OFF)
 -u: building unit test options(default: OFF)
 -l: run cpplint, clang-format and clang-tidy(default: OFF)
 -c: code coverage(default: OFF)
@@ -79,7 +74,7 @@ parameter:
 -h: help
 
 usage:
-./build.sh -o \${INSTALL_PREFIX} -t \${BUILD_TYPE} -b \${CORE_BUILD_DIR} -f \${FAISS_ROOT} [-p] [-g] [-x] [-u] [-l] [-c] [-m] [-h]
+./build.sh -o \${INSTALL_PREFIX} -t \${BUILD_TYPE} -b \${CORE_BUILD_DIR} -f \${FAISS_ROOT} [-p] [-g] [-u] [-l] [-c] [-m] [-h]
                 "
                 exit 0
                 ;;
@@ -104,7 +99,6 @@ CMAKE_CMD="cmake \
 -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
 -DCMAKE_CUDA_COMPILER=${CUDA_COMPILER} \
 -DMILVUS_GPU_VERSION=${GPU_VERSION} \
--DCUSTOMIZATION=${CUSTOMIZATION} \
 -DBUILD_UNIT_TEST=${BUILD_UNITTEST} \
 -DBUILD_COVERAGE=${BUILD_COVERAGE} \
 -DFAISS_ROOT=${FAISS_ROOT} \

core/CMakeLists.txt

@@ -160,7 +160,6 @@ if (MILVUS_USE_CCACHE)
 endif ()
 
 if (CUSTOMIZATION)
-    set(MILVUS_GPU_VERSION ON)
     add_compile_definitions(CUSTOMIZATION)
 endif ()
 

core/build.sh

@@ -9,7 +9,6 @@ BUILD_COVERAGE="OFF"
 DB_PATH="/tmp/milvus"
 PROFILING="OFF"
 RUN_CPPLINT="OFF"
-CUSTOMIZATION="OFF" # default use ori faiss
 CUDA_COMPILER=/usr/local/cuda/bin/nvcc
 GPU_VERSION="OFF" #defaults to CPU version
 WITH_MKL="OFF"
@@ -18,7 +17,7 @@ FAISS_SOURCE="BUNDLED"
 WITH_PROMETHEUS="ON"
 FIU_ENABLE="OFF"
 
-while getopts "p:d:t:f:ulrcghxzmei" arg; do
+while getopts "p:d:t:f:ulrcghzmei" arg; do
   case $arg in
   p)
     INSTALL_PREFIX=$OPTARG
@@ -52,9 +51,6 @@ while getopts "p:d:t:f:ulrcghxzmei" arg; do
   z)
     PROFILING="ON"
     ;;
-  x)
-    CUSTOMIZATION="ON"
-    ;;
   g)
     GPU_VERSION="ON"
     ;;
@@ -120,7 +116,6 @@ CMAKE_CMD="cmake \
 -DBUILD_COVERAGE=${BUILD_COVERAGE} \
 -DMILVUS_DB_PATH=${DB_PATH} \
 -DMILVUS_ENABLE_PROFILING=${PROFILING} \
--DCUSTOMIZATION=${CUSTOMIZATION} \
 -DMILVUS_GPU_VERSION=${GPU_VERSION} \
 -DFAISS_WITH_MKL=${WITH_MKL} \
 -DMILVUS_WITH_PROMETHEUS=${WITH_PROMETHEUS} \

core/cmake/DefineOptions.cmake

@@ -45,7 +45,7 @@ set_option_category("Milvus Build Option")
 
 define_option(MILVUS_GPU_VERSION "Build GPU version" OFF)
 
-define_option(CUSTOMIZATION "Build with customized FAISS library" OFF)
+define_option(CUSTOMIZATION "Build with customized FAISS library" ON)
 
 #----------------------------------------------------------------------
 set_option_category("Thirdparty")

core/src/db/engine/ExecutionEngineImpl.cpp

@@ -120,6 +120,7 @@ ExecutionEngineImpl::CreatetVecIndex(EngineType type) {
             break;
         }
 #ifdef CUSTOMIZATION
+#ifdef MILVUS_GPU_VERSION
         case EngineType::FAISS_IVFSQ8H: {
             if (gpu_resource_enable) {
                 index = GetVecIndexFactory(IndexType::FAISS_IVFSQ8_HYBRID);
@@ -128,6 +129,7 @@ ExecutionEngineImpl::CreatetVecIndex(EngineType type) {
             }
             break;
         }
+#endif
 #endif
         case EngineType::FAISS_PQ: {
 #ifdef MILVUS_GPU_VERSION

core/src/index/CMakeLists.txt

@@ -73,7 +73,6 @@ include(DefineOptionsCore)
 include(BuildUtilsCore)
 
 if (CUSTOMIZATION)
-    set(MILVUS_GPU_VERSION ON)
     add_compile_definitions(CUSTOMIZATION)
 endif ()
 

core/src/index/cmake/DefineOptionsCore.cmake

@@ -49,7 +49,7 @@ else ()
     define_option(KNOWHERE_GPU_VERSION "Build GPU version" OFF)
 endif ()
 
-define_option(CUSTOMIZATION "Build with customized FAISS library" OFF)
+define_option(CUSTOMIZATION "Build with customized FAISS library" ON)
 
 #----------------------------------------------------------------------
 set_option_category("Thirdparty")

core/src/index/cmake/ThirdPartyPackagesCore.cmake

@@ -206,11 +206,6 @@ foreach (_VERSION_ENTRY ${TOOLCHAIN_VERSIONS_TXT})
 endforeach ()
 
 set(FAISS_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/thirdparty/faiss)
-if (DEFINED ENV{FAISS_SOURCE_URL})
-    set(FAISS_SOURCE_URL "$ENV{FAISS_SOURCE_URL}")
-else ()
-    set(FAISS_SOURCE_URL "https://github.com/JinHai-CN/faiss/archive/${FAISS_VERSION}.tar.gz")
-endif ()
 
 if (DEFINED ENV{KNOWHERE_ARROW_URL})
     set(ARROW_SOURCE_URL "$ENV{KNOWHERE_ARROW_URL}")
@@ -467,15 +462,16 @@ macro(build_faiss)
                 "--with-cuda-arch=-gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75"
                 )
     else ()
-        set(FAISS_CONFIGURE_ARGS ${FAISS_CONFIGURE_ARGS} --without-cuda)
+        set(FAISS_CONFIGURE_ARGS ${FAISS_CONFIGURE_ARGS}
+                "CPPFLAGS=-DUSE_CPU"
+                --without-cuda)
     endif ()
 
-    if (CUSTOMIZATION)
+    if (DEFINED ENV{FAISS_SOURCE_URL})
+        set(FAISS_SOURCE_URL "$ENV{FAISS_SOURCE_URL}")
         externalproject_add(faiss_ep
-                DOWNLOAD_COMMAND
-                ""
-                SOURCE_DIR
-                ${FAISS_SOURCE_DIR}
+                URL
+                ${FAISS_SOURCE_URL}
                 ${EP_LOG_OPTIONS}
                 CONFIGURE_COMMAND
                 "./configure"
@@ -490,8 +486,10 @@ macro(build_faiss)
                 ${FAISS_STATIC_LIB})
     else ()
         externalproject_add(faiss_ep
-                URL
-                ${FAISS_SOURCE_URL}
+                DOWNLOAD_COMMAND
+                ""
+                SOURCE_DIR
+                ${FAISS_SOURCE_DIR}
                 ${EP_LOG_OPTIONS}
                 CONFIGURE_COMMAND
                 "./configure"

core/src/index/thirdparty/faiss/InvertedLists.cpp

@@ -14,6 +14,8 @@
 
 #include <faiss/utils/utils.h>
 #include <faiss/impl/FaissAssert.h>
+
+#ifndef USE_CPU
 #include "gpu/utils/DeviceUtils.h"
 #include "cuda.h"
 #include "cuda_runtime.h"
@@ -47,6 +49,7 @@ PageLockMemory::PageLockMemory(PageLockMemory &&other) {
     other.nbytes = 0;
 }
 }
+#endif
 
 namespace faiss {
 
@@ -257,6 +260,11 @@ ReadOnlyArrayInvertedLists::ReadOnlyArrayInvertedLists(size_t nlist,
     auto total_size = std::accumulate(readonly_length.begin(), readonly_length.end(), 0);
     readonly_offset.reserve(nlist);
 
+#ifdef USE_CPU
+    readonly_codes.reserve(total_size * code_size);
+    readonly_ids.reserve(total_size);
+#endif
+
     size_t offset = 0;
     for (auto i=0; i<readonly_length.size(); ++i) {
         readonly_offset.emplace_back(offset);
@@ -266,8 +274,10 @@ ReadOnlyArrayInvertedLists::ReadOnlyArrayInvertedLists(size_t nlist,
 
 ReadOnlyArrayInvertedLists::ReadOnlyArrayInvertedLists(const ArrayInvertedLists& other)
         : InvertedLists (other.nlist, other.code_size) {
+#ifndef USE_CPU
     std::vector <uint8_t> readonly_codes;
     std::vector <idx_t> readonly_ids;
+#endif
     readonly_length.reserve(nlist);
     size_t offset = 0;
     for (auto& list_ids : other.ids) {
@@ -281,6 +291,7 @@ ReadOnlyArrayInvertedLists::ReadOnlyArrayInvertedLists(const ArrayInvertedLists&
         readonly_codes.insert(readonly_codes.end(), list_codes.begin(), list_codes.end());
     }
 
+#ifndef USE_CPU
     // convert to page-lock memory
     {
         size_t size = readonly_codes.size() * sizeof(uint8_t);
@@ -292,6 +303,7 @@ ReadOnlyArrayInvertedLists::ReadOnlyArrayInvertedLists(const ArrayInvertedLists&
         pin_readonly_ids = std::make_shared<PageLockMemory>(size);
         memcpy(pin_readonly_ids->data, readonly_ids.data(), size);
     }
+#endif
 
     valid = true;
 }
@@ -352,25 +364,41 @@ size_t ReadOnlyArrayInvertedLists::list_size(size_t list_no) const
 const uint8_t * ReadOnlyArrayInvertedLists::get_codes (size_t list_no) const
 {
     FAISS_ASSERT(list_no < nlist && valid);
+#ifdef USE_CPU
+    return readonly_codes.data() + readonly_offset[list_no] * code_size;
+#else
     uint8_t *pcodes = (uint8_t *)(pin_readonly_codes->data);
     return pcodes + readonly_offset[list_no] * code_size;
+#endif
 }
 
 const InvertedLists::idx_t* ReadOnlyArrayInvertedLists::get_ids (size_t list_no) const
 {
     FAISS_ASSERT(list_no < nlist && valid);
+#ifdef USE_CPU
+    return readonly_ids.data() + readonly_offset[list_no];
+#else
     idx_t *pids = (idx_t *)pin_readonly_ids->data;
     return pids + readonly_offset[list_no];
+#endif
 }
 
 const InvertedLists::idx_t* ReadOnlyArrayInvertedLists::get_all_ids() const {
     FAISS_ASSERT(valid);
+#ifdef USE_CPU
+    return readonly_ids.data();
+#else
     return (idx_t *)(pin_readonly_ids->data);
+#endif
 }
 
 const uint8_t* ReadOnlyArrayInvertedLists::get_all_codes() const {
     FAISS_ASSERT(valid);
+#ifdef USE_CPU
+    return readonly_codes.data();
+#else
     return (uint8_t *)(pin_readonly_codes->data);
+#endif
 }
 
 const std::vector<size_t>& ReadOnlyArrayInvertedLists::get_list_length() const {

core/src/index/thirdparty/faiss/InvertedLists.h

@@ -19,6 +19,8 @@
 #include <vector>
 #include <faiss/Index.h>
 
+
+#ifndef USE_CPU
 namespace faiss {
 
 struct PageLockMemory {
@@ -42,6 +44,7 @@ public:
 };
 using PageLockMemoryPtr = std::shared_ptr<PageLockMemory>;
 }
+#endif
 
 namespace faiss {
 
@@ -232,10 +235,14 @@ struct ArrayInvertedLists: InvertedLists {
 };
 
 struct ReadOnlyArrayInvertedLists: InvertedLists {
+#ifdef USE_CPU
+    std::vector <uint8_t> readonly_codes;
+    std::vector <idx_t> readonly_ids;
+#else
     PageLockMemoryPtr pin_readonly_codes;
     PageLockMemoryPtr pin_readonly_ids;
-//    std::vector <uint8_t> readonly_codes;
-//    std::vector <idx_t> readonly_ids;
+#endif
+
     std::vector <size_t> readonly_length;
     std::vector <size_t> readonly_offset;
     bool valid;

core/src/index/thirdparty/faiss/impl/index_read.cpp

@@ -211,12 +211,17 @@ InvertedLists *read_InvertedLists (IOReader *f, int io_flags) {
         auto ails = new ReadOnlyArrayInvertedLists(nlist, code_size, list_length);
         size_t n;
         READ1(n);
-//        ails->readonly_ids.resize(n);
-//        ails->readonly_codes.resize(n*code_size);
+#ifdef USE_CPU
+        ails->readonly_ids.resize(n);
+        ails->readonly_codes.resize(n*code_size);
+        READANDCHECK(ails->readonly_ids.data(), n);
+        READANDCHECK(ails->readonly_codes.data(), n * code_size);
+#else
         ails->pin_readonly_ids = std::make_shared<PageLockMemory>(n * sizeof(InvertedLists::idx_t));
         ails->pin_readonly_codes = std::make_shared<PageLockMemory>(n * code_size * sizeof(uint8_t));
         READANDCHECK((InvertedLists::idx_t *) ails->pin_readonly_ids->data, n);
         READANDCHECK((uint8_t *) ails->pin_readonly_codes->data, n * code_size);
+#endif
         return ails;
     } else if (h == fourcc ("ilar") && !(io_flags & IO_FLAG_MMAP)) {
         auto ails = new ArrayInvertedLists (0, 0);

core/src/index/thirdparty/faiss/impl/index_write.cpp

@@ -245,12 +245,17 @@ void write_InvertedLists (const InvertedLists *ils, IOWriter *f) {
         WRITE1 (oa->nlist);
         WRITE1 (oa->code_size);
         WRITEVECTOR(oa->readonly_length);
+#ifdef USE_CPU
+        size_t n = oa->readonly_ids.size();
+        WRITE1(n);
+        WRITEANDCHECK(oa->readonly_ids.data(), n);
+        WRITEANDCHECK(oa->readonly_codes.data(), n * oa->code_size);
+#else
         size_t n = oa->pin_readonly_ids->size() / sizeof(InvertedLists::idx_t);
         WRITE1(n);
-//        WRITEANDCHECK(oa->readonly_ids.data(), n);
-//        WRITEANDCHECK(oa->readonly_codes.data(), n * oa->code_size);
         WRITEANDCHECK((InvertedLists::idx_t *) oa->pin_readonly_ids->data, n);
         WRITEANDCHECK((uint8_t *) oa->pin_readonly_codes->data, n * oa->code_size);
+#endif
     } else if (const auto & od =
                dynamic_cast<const OnDiskInvertedLists *>(ils)) {
         uint32_t h = fourcc ("ilod");

core/src/index/thirdparty/versions.txt

@@ -3,5 +3,4 @@ BOOST_VERSION=1.70.0
 GTEST_VERSION=1.8.1
 LAPACK_VERSION=v3.8.0
 OPENBLAS_VERSION=v0.3.6
-FAISS_VERSION=1.6.0
 MKL_VERSION=2019.5.281

core/src/wrapper/VecIndex.cpp

@@ -170,6 +170,7 @@ GetVecIndexFactory(const IndexType& type, const Config& cfg) {
 
 #endif
 #ifdef CUSTOMIZATION
+#ifdef MILVUS_GPU_VERSION
         case IndexType::FAISS_IVFSQ8_HYBRID: {
             server::Config& config = server::Config::GetInstance();
             bool gpu_resource_enable = true;
@@ -181,6 +182,7 @@ GetVecIndexFactory(const IndexType& type, const Config& cfg) {
                 throw Exception(DB_ERROR, "No GPU resources for IndexType::FAISS_IVFSQ8_HYBRID");
             }
         }
+#endif
 #endif
         case IndexType::NSG_MIX: {  // TODO(linxj): bug.
             index = std::make_shared<knowhere::NSG>(gpu_device);

core/unittest/db/test_db.cpp

@@ -510,9 +510,11 @@ TEST_F(DBTest, INDEX_TEST) {
     ASSERT_TRUE(stat.ok());
 
 #ifdef CUSTOMIZATION
+#ifdef MILVUS_GPU_VERSION
     index.engine_type_ = (int)milvus::engine::EngineType::FAISS_IVFSQ8H;
     stat = db_->CreateIndex(table_info.table_id_, index);
     ASSERT_TRUE(stat.ok());
+#endif
 #endif
 
     milvus::engine::TableIndex index_out;