IGNORE: Test PR for AMX build in CI

.github/actions/build_cmake/action.yml

@@ -4,6 +4,10 @@ inputs:
     description: 'Compile options / optimization level.'
     required: false
     default: generic
+  dnnl:
+    description: 'Enable DNNL support.'
+    required: false
+    default: OFF
   gpu:
     description: 'Enable GPU support.'
     required: false
@@ -53,6 +57,11 @@ runs:
           conda install -y -q libraft cuda-version=11.8 cuda-toolkit -c rapidsai-nightly -c "nvidia/label/cuda-11.8.0" -c conda-forge
         fi
 
+        if [ "${{ inputs.dnnl }}" = "ON" ]; then
+          # resurrect
+          conda install -y -q conda-forge::onednn
+        fi
+
         # install test packages
         conda install -y pytest
         if [ "${{ inputs.gpu }}" = "ON" ]; then
@@ -70,6 +79,7 @@ runs:
           cmake -B build \
                 -DBUILD_TESTING=ON \
                 -DBUILD_SHARED_LIBS=ON  \
+                -DFAISS_ENABLE_DNNL=${{ inputs.dnnl }} \
                 -DFAISS_ENABLE_GPU=${{ inputs.gpu }}  \
                 -DFAISS_ENABLE_RAFT=${{ inputs.raft }}  \
                 -DFAISS_OPT_LEVEL=${{ inputs.opt_level }}  \

.github/workflows/build.yml

@@ -11,249 +11,12 @@ env:
   OMP_NUM_THREADS: '10'
   MKL_THREADING_LAYER: GNU
 jobs:
-  format:
-    name: Format
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Install clang-format
-        run: |
-            sudo apt-get update -y
-            sudo apt-get install -y wget
-            sudo apt install -y lsb-release wget software-properties-common gnupg
-            wget https://apt.llvm.org/llvm.sh
-            chmod u+x llvm.sh
-            sudo ./llvm.sh 18
-            sudo apt-get install -y git-core clang-format-18
-      - name: Verify clang-format
-        run: |
-            git ls-files | grep -E  '\.(cpp|h|cu|cuh)$' | xargs clang-format-18 -i
-            if git diff --quiet; then
-              echo "Formatting OK!"
-            else
-              echo "Formatting not OK!"
-              echo "------------------"
-              git --no-pager diff --color
-              exit 1
-            fi
-  linux-x86_64-cmake:
-    name: Linux x86_64 (cmake)
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - uses: ./.github/actions/build_cmake
-  linux-x86_64-AVX2-cmake:
-    name: Linux x86_64 AVX2 (cmake)
-    needs: linux-x86_64-cmake
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - uses: ./.github/actions/build_cmake
-        with:
-          opt_level: avx2
-  linux-x86_64-AVX512-cmake:
-    name: Linux x86_64 AVX512 (cmake)
-    continue-on-error: true # non-blocking mode for now
-    needs: linux-x86_64-cmake
+  AMX-DNNL-cmake:
+    name: AMX DNNL (cmake)
     runs-on: faiss-aws-m7i.large
-    steps:
-      - name: Checkout
-        continue-on-error: true # non-blocking mode for now
-        uses: actions/checkout@v4
-      - uses: ./.github/actions/build_cmake
-        with:
-          opt_level: avx512
-        continue-on-error: true # non-blocking mode for now
-  linux-x86_64-GPU-cmake:
-    name: Linux x86_64 GPU (cmake)
-    needs: linux-x86_64-cmake
-    runs-on: 4-core-ubuntu-gpu-t4
     steps:
       - name: Checkout
         uses: actions/checkout@v4
       - uses: ./.github/actions/build_cmake
         with:
-          gpu: ON
-  linux-x86_64-GPU-w-RAFT-cmake:
-    name: Linux x86_64 GPU w/ RAFT (cmake)
-    needs: linux-x86_64-cmake
-    runs-on: 4-core-ubuntu-gpu-t4
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - uses: ./.github/actions/build_cmake
-        with:
-          gpu: ON
-          raft: ON
-  linux-arm64-SVE-cmake:
-    name: Linux arm64 SVE (cmake)
-    runs-on: faiss-aws-r8g.large
-    continue-on-error: true # non-blocking mode for now
-    steps:
-      - name: Checkout
-        continue-on-error: true # non-blocking mode for now
-        uses: actions/checkout@v4
-      - uses: ./.github/actions/build_cmake
-        continue-on-error: true # non-blocking mode for now
-      # TODO(T197096427): uncomment this once SVE PR is merged
-      #   with:
-      #     opt_level: sve
-  linux-x86_64-conda:
-    name: Linux x86_64 (conda)
-    needs: linux-x86_64-cmake
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-  windows-x86_64-conda:
-    name: Windows x86_64 (conda)
-    needs: linux-x86_64-cmake
-    runs-on: windows-2019
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-  linux-arm64-conda:
-    name: Linux arm64 (conda)
-    needs: linux-x86_64-cmake
-    runs-on: 2-core-ubuntu-arm
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-  linux-x86_64-packages:
-    name: Linux x86_64 packages
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-  linux-x86_64-GPU-packages-CUDA-11-4-4:
-    name: Linux x86_64 GPU packages (CUDA 11.4.4)
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: 4-core-ubuntu-gpu-t4
-    env:
-      CUDA_ARCHS: "60-real;61-real;62-real;70-real;72-real;75-real;80;86-real"
-      FAISS_FLATTEN_CONDA_INCLUDES: "1"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-          cuda: "11.4.4"
-          compiler_version: "11.2"
-  linux-x86_64-GPU-RAFT-packages-CUDA11-8-0:
-    name: Linux x86_64 GPU w/ RAFT packages (CUDA 11.8.0)
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: 4-core-ubuntu-gpu-t4
-    env:
-      CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-          raft: "ON"
-          cuda: "11.8.0"
-          compiler_version: "11.2"
-  linux-x86_64-GPU-packages-CUDA-12-1-1:
-    name: Linux x86_64 GPU packages (CUDA 12.1.1)
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: 4-core-ubuntu-gpu-t4
-    env:
-      CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-          cuda: "12.1.1"
-          compiler_version: "11.2"
-  linux-x86_64-GPU-RAFT-packages-CUDA12-1-1:
-    name: Linux x86_64 GPU w/ RAFT packages (CUDA 12.1.1)
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: 4-core-ubuntu-gpu-t4
-    env:
-      CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-          raft: "ON"
-          cuda: "12.1.1"
-          compiler_version: "11.2"
-  windows-x86_64-packages:
-    name: Windows x86_64 packages
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: windows-2019
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-  osx-arm64-packages:
-    name: OSX arm64 packages
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: macos-14
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
-  linux-arm64-packages:
-    name: Linux arm64 packages
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
-    runs-on: 2-core-ubuntu-arm
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          fetch-tags: true
-      - uses: ./.github/actions/build_conda
-        with:
-          label: main
+          dnnl: ON

CMakeLists.txt

@@ -52,6 +52,7 @@ list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake")
 
 # Valid values are "generic", "avx2", "avx512".
 option(FAISS_OPT_LEVEL "" "generic")
+option(FAISS_ENABLE_DNNL "Enable support for onednn to accelerate indexflat search." OFF)
 option(FAISS_ENABLE_GPU "Enable support for GPU indexes." ON)
 option(FAISS_ENABLE_RAFT "Enable RAFT for GPU indexes." OFF)
 option(FAISS_ENABLE_PYTHON "Build Python extension." ON)
@@ -80,6 +81,7 @@ if(FAISS_ENABLE_C_API)
   add_subdirectory(c_api)
 endif()
 
+
 add_subdirectory(demos)
 add_subdirectory(benchs)
 add_subdirectory(tutorial/cpp)

INSTALL.md

@@ -87,6 +87,9 @@ The optional requirements are:
 - for GPU indices:
   - nvcc,
   - the CUDA toolkit,
+- for Intel®-AMX/oneDNN acceleration:
+  - oneDNN,
+  - 4th+ Gen Intel® Xeon® Scalable processor.
 - for the python bindings:
   - python 3,
   - numpy,
@@ -108,6 +111,8 @@ Several options can be passed to CMake, among which:
 - general options:
   - `-DFAISS_ENABLE_GPU=OFF` in order to disable building GPU indices (possible
   values are `ON` and `OFF`),
+  - `-DFAISS_ENABLE_DNNL=OFF` in order to support for Intel®-AMX/oneDNN to accelerate indexflat(inner_product) search (possible
+  values are `ON` and `OFF`, before invoking CMake and setting this option to `ON`, you can refer to this [link](https://oneapi-src.github.io/oneDNN/dev_guide_build.html) for installing oneDNN),
   - `-DFAISS_ENABLE_PYTHON=OFF` in order to disable building python bindings
   (possible values are `ON` and `OFF`),
   - `-DFAISS_ENABLE_RAFT=ON` in order to enable building the RAFT implementations

c_api/utils/distances_c.cpp

@@ -100,3 +100,11 @@ void faiss_set_distance_compute_min_k_reservoir(int value) {
 int faiss_get_distance_compute_min_k_reservoir() {
     return faiss::distance_compute_min_k_reservoir;
 }
+
+void faiss_set_distance_compute_dnnl_query_bs(int value) {
+    faiss::distance_compute_dnnl_query_bs = value;
+}
+
+int faiss_get_distance_compute_dnnl_query_bs() {
+    return faiss::distance_compute_dnnl_query_bs;
+}

c_api/utils/distances_c.h

@@ -103,6 +103,18 @@ void faiss_set_distance_compute_min_k_reservoir(int value);
 /// rather than a heap
 int faiss_get_distance_compute_min_k_reservoir();
 
+/// Setter of block sizes value for oneDNN/AMX distance computations
+void faiss_set_distance_compute_dnnl_query_bs(int value);
+
+/// Getter of block sizes value for oneDNN/AMX distance computations
+int faiss_get_distance_compute_dnnl_query_bs();
+
+/// Setter of block sizes value for oneDNN/AMX distance computations
+void faiss_set_distance_compute_dnnl_database_bs(int value);
+
+/// Getter of block sizes value for oneDNN/AMX distance computations
+int faiss_get_distance_compute_dnnl_database_bs();
+
 #ifdef __cplusplus
 }
 #endif

faiss/CMakeLists.txt

@@ -229,6 +229,15 @@ if(NOT WIN32)
   list(APPEND FAISS_HEADERS invlists/OnDiskInvertedLists.h)
 endif()
 
+if(FAISS_ENABLE_DNNL)
+  list(APPEND FAISS_HEADERS utils/onednn/onednn_utils.h)
+endif()
+
+if(FAISS_ENABLE_DNNL)
+   add_compile_definitions(ENABLE_DNNL)
+endif()
+
+
 # Export FAISS_HEADERS variable to parent scope.
 set(FAISS_HEADERS ${FAISS_HEADERS} PARENT_SCOPE)
 
@@ -300,6 +309,15 @@ endif()
 target_compile_definitions(faiss_avx2 PRIVATE FINTEGER=int)
 target_compile_definitions(faiss_avx512 PRIVATE FINTEGER=int)
 
+if(FAISS_ENABLE_DNNL)
+  find_library(RT_LIB rt)
+  find_library(DNNL_LIB dnnl)
+  message(DNNL_LIB=${DNNL_LIB})
+  target_link_libraries(faiss PRIVATE ${RT_LIB} ${DNNL_LIB})
+  target_link_libraries(faiss_avx2 PRIVATE ${RT_LIB} ${DNNL_LIB})
+  target_link_libraries(faiss_avx512 PRIVATE ${RT_LIB} ${DNNL_LIB})
+endif()
+
 find_package(OpenMP REQUIRED)
 target_link_libraries(faiss PRIVATE OpenMP::OpenMP_CXX)
 target_link_libraries(faiss_avx2 PRIVATE OpenMP::OpenMP_CXX)