Introduce avx512 optimization mode and FAISS_OPT_LEVEL env variable

Signed-off-by: Alexandr Guzhva <[email protected]>

Author: alexanderguzhva

CMakeLists.txt

@@ -50,7 +50,7 @@ set(CMAKE_CXX_STANDARD 17)
 
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake")
 
-# Valid values are "generic", "avx2".
+# Valid values are "generic", "avx2", "avx512".
 option(FAISS_OPT_LEVEL "" "generic")
 option(FAISS_ENABLE_GPU "Enable support for GPU indexes." ON)
 option(FAISS_ENABLE_RAFT "Enable RAFT for GPU indexes." OFF)

INSTALL.md

@@ -119,8 +119,8 @@ Several options can be passed to CMake, among which:
   - `-DCMAKE_BUILD_TYPE=Release` in order to enable generic compiler
   optimization options (enables `-O3` on gcc for instance),
   - `-DFAISS_OPT_LEVEL=avx2` in order to enable the required compiler flags to
-  generate code using optimized SIMD instructions (possible values are `generic`
-  and `avx2`, by increasing order of optimization),
+  generate code using optimized SIMD instructions (possible values are `generic`,
+  `avx2` and `avx512`, by increasing order of optimization),
 - BLAS-related options:
   - `-DBLA_VENDOR=Intel10_64_dyn -DMKL_LIBRARIES=/path/to/mkl/libs` to use the
   Intel MKL BLAS implementation, which is significantly faster than OpenBLAS

faiss/CMakeLists.txt

@@ -244,12 +244,29 @@ else()
   add_compile_options(/bigobj)
 endif()
 
+add_library(faiss_avx512 ${FAISS_SRC})
+if(NOT FAISS_OPT_LEVEL STREQUAL "avx512")
+  set_target_properties(faiss_avx512 PROPERTIES EXCLUDE_FROM_ALL TRUE)
+endif()
+if(NOT WIN32)
+  # All modern CPUs support F, CD, VL, DQ, BW extensions.
+  # Ref: https://en.wikipedia.org/wiki/AVX512
+  target_compile_options(faiss_avx512 PRIVATE $<$<COMPILE_LANGUAGE:CXX>:-mavx2 -mfma -mf16c -mavx512f -mavx512cd -mavx512vl -mavx512dq -mavx512bw -mpopcnt>)
+else()
+  target_compile_options(faiss_avx512 PRIVATE $<$<COMPILE_LANGUAGE:CXX>:/arch:AVX512>)
+  # we need bigobj for the swig wrapper
+  add_compile_options(/bigobj)
+endif()
+
 # Handle `#include <faiss/foo.h>`.
 target_include_directories(faiss PUBLIC
   $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}>)
 # Handle `#include <faiss/foo.h>`.
 target_include_directories(faiss_avx2 PUBLIC
   $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}>)
+# Handle `#include <faiss/foo.h>`.
+target_include_directories(faiss_avx512 PUBLIC
+  $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}>)
 
 set_target_properties(faiss PROPERTIES
   POSITION_INDEPENDENT_CODE ON
@@ -259,31 +276,41 @@ set_target_properties(faiss_avx2 PROPERTIES
   POSITION_INDEPENDENT_CODE ON
   WINDOWS_EXPORT_ALL_SYMBOLS ON
 )
+set_target_properties(faiss_avx512 PROPERTIES
+  POSITION_INDEPENDENT_CODE ON
+  WINDOWS_EXPORT_ALL_SYMBOLS ON
+)
 
 if(WIN32)
   target_compile_definitions(faiss PRIVATE FAISS_MAIN_LIB)
   target_compile_definitions(faiss_avx2 PRIVATE FAISS_MAIN_LIB)
+  target_compile_definitions(faiss_avx512 PRIVATE FAISS_MAIN_LIB)
 endif()
 
 target_compile_definitions(faiss PRIVATE FINTEGER=int)
 target_compile_definitions(faiss_avx2 PRIVATE FINTEGER=int)
+target_compile_definitions(faiss_avx512 PRIVATE FINTEGER=int)
 
 find_package(OpenMP REQUIRED)
 target_link_libraries(faiss PRIVATE OpenMP::OpenMP_CXX)
 target_link_libraries(faiss_avx2 PRIVATE OpenMP::OpenMP_CXX)
+target_link_libraries(faiss_avx512 PRIVATE OpenMP::OpenMP_CXX)
 
 find_package(MKL)
 if(MKL_FOUND)
   target_link_libraries(faiss PRIVATE ${MKL_LIBRARIES})
   target_link_libraries(faiss_avx2 PRIVATE ${MKL_LIBRARIES})
+  target_link_libraries(faiss_avx512 PRIVATE ${MKL_LIBRARIES})
 else()
   find_package(BLAS REQUIRED)
   target_link_libraries(faiss PRIVATE ${BLAS_LIBRARIES})
   target_link_libraries(faiss_avx2 PRIVATE ${BLAS_LIBRARIES})
+  target_link_libraries(faiss_avx512 PRIVATE ${BLAS_LIBRARIES})
 
   find_package(LAPACK REQUIRED)
   target_link_libraries(faiss PRIVATE ${LAPACK_LIBRARIES})
   target_link_libraries(faiss_avx2 PRIVATE ${LAPACK_LIBRARIES})
+  target_link_libraries(faiss_avx512 PRIVATE ${LAPACK_LIBRARIES})
 endif()
 
 install(TARGETS faiss
@@ -300,6 +327,13 @@ if(FAISS_OPT_LEVEL STREQUAL "avx2")
     LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
   )
 endif()
+if(FAISS_OPT_LEVEL STREQUAL "avx512")
+  install(TARGETS faiss_avx512
+    EXPORT faiss-targets
+    ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+  )
+endif()
 
 foreach(header ${FAISS_HEADERS})
   get_filename_component(dir ${header} DIRECTORY )

faiss/python/CMakeLists.txt

@@ -44,20 +44,24 @@ endmacro()
 # CMake's SWIG wrappers only allow tweaking certain settings at source level, so
 # we duplicate the source in order to override the module name.
 configure_file(swigfaiss.swig ${CMAKE_CURRENT_SOURCE_DIR}/swigfaiss_avx2.swig COPYONLY)
+configure_file(swigfaiss.swig ${CMAKE_CURRENT_SOURCE_DIR}/swigfaiss_avx512.swig COPYONLY)
 
 configure_swigfaiss(swigfaiss.swig)
 configure_swigfaiss(swigfaiss_avx2.swig)
+configure_swigfaiss(swigfaiss_avx512.swig)
 
 if(TARGET faiss)
   # Manually add headers as extra dependencies of swigfaiss.
   set(SWIG_MODULE_swigfaiss_EXTRA_DEPS)
   foreach(h ${FAISS_HEADERS})
     list(APPEND SWIG_MODULE_swigfaiss_EXTRA_DEPS "${faiss_SOURCE_DIR}/faiss/${h}")
     list(APPEND SWIG_MODULE_swigfaiss_avx2_EXTRA_DEPS "${faiss_SOURCE_DIR}/faiss/${h}")
+    list(APPEND SWIG_MODULE_swigfaiss_avx512_EXTRA_DEPS "${faiss_SOURCE_DIR}/faiss/${h}")
   endforeach()
   foreach(h ${FAISS_GPU_HEADERS})
     list(APPEND SWIG_MODULE_swigfaiss_EXTRA_DEPS "${faiss_SOURCE_DIR}/faiss/gpu/${h}")
     list(APPEND SWIG_MODULE_swigfaiss_avx2_EXTRA_DEPS "${faiss_SOURCE_DIR}/faiss/gpu/${h}")
+    list(APPEND SWIG_MODULE_swigfaiss_avx512_EXTRA_DEPS "${faiss_SOURCE_DIR}/faiss/gpu/${h}")
   endforeach()
 else()
   find_package(faiss REQUIRED)
@@ -82,14 +86,28 @@ if(NOT FAISS_OPT_LEVEL STREQUAL "avx2")
   set_target_properties(swigfaiss_avx2 PROPERTIES EXCLUDE_FROM_ALL TRUE)
 endif()
 
+set_property(SOURCE swigfaiss_avx512.swig
+  PROPERTY SWIG_MODULE_NAME swigfaiss_avx512)
+swig_add_library(swigfaiss_avx512
+  TYPE SHARED
+  LANGUAGE python
+  SOURCES swigfaiss_avx512.swig
+)
+set_property(TARGET swigfaiss_avx512 PROPERTY SWIG_COMPILE_OPTIONS -doxygen)
+if(NOT FAISS_OPT_LEVEL STREQUAL "avx512")
+  set_target_properties(swigfaiss_avx512 PROPERTIES EXCLUDE_FROM_ALL TRUE)
+endif()
+
 if(NOT WIN32)
   # NOTE: Python does not recognize the dylib extension.
   set_target_properties(swigfaiss PROPERTIES SUFFIX .so)
   set_target_properties(swigfaiss_avx2 PROPERTIES SUFFIX .so)
+  set_target_properties(swigfaiss_avx512 PROPERTIES SUFFIX .so)
 else()
   # we need bigobj for the swig wrapper
   target_compile_options(swigfaiss PRIVATE /bigobj)
   target_compile_options(swigfaiss_avx2 PRIVATE /bigobj)
+  target_compile_options(swigfaiss_avx512 PRIVATE /bigobj)
 endif()
 
 if(FAISS_ENABLE_GPU)
@@ -99,6 +117,7 @@ if(FAISS_ENABLE_GPU)
   endif()
   target_link_libraries(swigfaiss PRIVATE CUDA::cudart $<$<BOOL:${FAISS_ENABLE_RAFT}>:raft::raft> $<$<BOOL:${FAISS_ENABLE_RAFT}>:nvidia::cutlass::cutlass>)
   target_link_libraries(swigfaiss_avx2 PRIVATE CUDA::cudart $<$<BOOL:${FAISS_ENABLE_RAFT}>:raft::raft> $<$<BOOL:${FAISS_ENABLE_RAFT}>:nvidia::cutlass::cutlass>)
+  target_link_libraries(swigfaiss_avx512 PRIVATE CUDA::cudart $<$<BOOL:${FAISS_ENABLE_RAFT}>:raft::raft> $<$<BOOL:${FAISS_ENABLE_RAFT}>:nvidia::cutlass::cutlass>)
 endif()
 
 find_package(OpenMP REQUIRED)
@@ -117,10 +136,18 @@ target_link_libraries(swigfaiss_avx2 PRIVATE
   OpenMP::OpenMP_CXX
 )
 
+target_link_libraries(swigfaiss_avx512 PRIVATE
+  faiss_avx512
+  Python::Module
+  Python::NumPy
+  OpenMP::OpenMP_CXX
+)
+
 # Hack so that python_callbacks.h can be included as
 # `#include <faiss/python/python_callbacks.h>`.
 target_include_directories(swigfaiss PRIVATE ${PROJECT_SOURCE_DIR}/../..)
 target_include_directories(swigfaiss_avx2 PRIVATE ${PROJECT_SOURCE_DIR}/../..)
+target_include_directories(swigfaiss_avx512 PRIVATE ${PROJECT_SOURCE_DIR}/../..)
 
 find_package(Python REQUIRED
   COMPONENTS Development NumPy
@@ -140,6 +167,7 @@ target_include_directories(faiss_python_callbacks PRIVATE ${Python_INCLUDE_DIRS}
 
 target_link_libraries(swigfaiss PRIVATE faiss_python_callbacks)
 target_link_libraries(swigfaiss_avx2 PRIVATE faiss_python_callbacks)
+target_link_libraries(swigfaiss_avx512 PRIVATE faiss_python_callbacks)
 
 configure_file(setup.py setup.py COPYONLY)
 configure_file(__init__.py __init__.py COPYONLY)

faiss/python/loader.py

@@ -18,7 +18,7 @@ def supported_instruction_sets():
 
     Example:
     >>> supported_instruction_sets()  # for x86
-    {"SSE2", "AVX2", ...}
+    {"SSE2", "AVX2", "AVX512", ...}
     >>> supported_instruction_sets()  # for PPC
     {"VSX", "VSX2", ...}
     >>> supported_instruction_sets()  # for ARM
@@ -41,25 +41,58 @@ def supported_instruction_sets():
             return {"AVX2"}
     elif platform.system() == "Linux":
         import numpy.distutils.cpuinfo
+        result = set()
         if "avx2" in numpy.distutils.cpuinfo.cpu.info[0].get('flags', ""):
-            return {"AVX2"}
+            result.add("AVX2")
+        if "avx512" in numpy.distutils.cpuinfo.cpu.info[0].get('flags', ""):
+            result.add("AVX512")
+        return result
     return set()
 
 
 logger = logging.getLogger(__name__)
 
-has_AVX2 = "AVX2" in supported_instruction_sets()
-if has_AVX2:
+instruction_sets = None
+
+# try to load optimization level from env variable
+opt_env_variable_name = "FAISS_OPT_LEVEL"
+opt_level = os.environ.get(opt_env_variable_name, None)
+
+if opt_level is None:
+    logger.debug(f"Environment variable {opt_env_variable_name} is not set, " \
+                "so let's pick the instruction set according to the current CPU")
+    instruction_sets = supported_instruction_sets()
+else:
+    logger.debug(f"Using {opt_level} as an instruction set.")
+    instruction_sets = set()
+    instruction_sets.add(opt_level)
+
+loaded = False
+has_AVX512 = any("AVX512" in x.upper() for x in instruction_sets)
+if has_AVX512:
+    try:
+        logger.info("Loading faiss with AVX512 support.")
+        from .swigfaiss_avx512 import *
+        logger.info("Successfully loaded faiss with AVX512 support.")
+        loaded = True
+    except ImportError as e:
+        logger.info(f"Could not load library with AVX512 support due to:\n{e!r}")
+        # reset so that we load without AVX512 below
+        loaded = False
+
+has_AVX2 = "AVX2" in instruction_sets
+if has_AVX2 and not loaded:
     try:
         logger.info("Loading faiss with AVX2 support.")
         from .swigfaiss_avx2 import *
         logger.info("Successfully loaded faiss with AVX2 support.")
+        loaded = True
     except ImportError as e:
         logger.info(f"Could not load library with AVX2 support due to:\n{e!r}")
         # reset so that we load without AVX2 below
-        has_AVX2 = False
+        loaded = False
 
-if not has_AVX2:
+if not loaded:
     # we import * so that the symbol X can be accessed as faiss.X
     logger.info("Loading faiss.")
     from .swigfaiss import *

faiss/python/setup.py

@@ -25,13 +25,15 @@
 
 swigfaiss_generic_lib = f"{prefix}_swigfaiss{ext}"
 swigfaiss_avx2_lib = f"{prefix}_swigfaiss_avx2{ext}"
+swigfaiss_avx512_lib = f"{prefix}_swigfaiss_avx512{ext}"
 
 found_swigfaiss_generic = os.path.exists(swigfaiss_generic_lib)
 found_swigfaiss_avx2 = os.path.exists(swigfaiss_avx2_lib)
+found_swigfaiss_avx512 = os.path.exists(swigfaiss_avx512_lib)
 
-assert (found_swigfaiss_generic or found_swigfaiss_avx2), \
+assert (found_swigfaiss_generic or found_swigfaiss_avx2 or found_swigfaiss_avx512), \
     f"Could not find {swigfaiss_generic_lib} or " \
-    f"{swigfaiss_avx2_lib}. Faiss may not be compiled yet."
+    f"{swigfaiss_avx2_lib} or {swigfaiss_avx512_lib}. Faiss may not be compiled yet."
 
 if found_swigfaiss_generic:
     print(f"Copying {swigfaiss_generic_lib}")
@@ -43,6 +45,11 @@
     shutil.copyfile("swigfaiss_avx2.py", "faiss/swigfaiss_avx2.py")
     shutil.copyfile(swigfaiss_avx2_lib, f"faiss/_swigfaiss_avx2{ext}")
 
+if found_swigfaiss_avx512:
+    print(f"Copying {swigfaiss_avx512_lib}")
+    shutil.copyfile("swigfaiss_avx512.py", "faiss/swigfaiss_avx512.py")
+    shutil.copyfile(swigfaiss_avx512_lib, f"faiss/_swigfaiss_avx512{ext}")
+
 long_description="""
 Faiss is a library for efficient similarity search and clustering of dense
 vectors. It contains algorithms that search in sets of vectors of any size,

faiss/utils/distances_fused/avx512.cpp

@@ -9,7 +9,7 @@
 
 #include <faiss/utils/distances_fused/avx512.h>
 
-#ifdef __AVX512__
+#ifdef __AVX512F__
 
 #include <immintrin.h>
 

faiss/utils/distances_fused/avx512.h

@@ -16,7 +16,7 @@
 
 #include <faiss/utils/Heap.h>
 
-#ifdef __AVX512__
+#ifdef __AVX512F__
 
 namespace faiss {
 

faiss/utils/distances_fused/distances_fused.cpp

@@ -27,7 +27,7 @@ bool exhaustive_L2sqr_fused_cmax(
         return true;
     }
 
-#ifdef __AVX512__
+#ifdef __AVX512F__
     // avx512 kernel
     return exhaustive_L2sqr_fused_cmax_AVX512(x, y, d, nx, ny, res, y_norms);
 #elif defined(__AVX2__) || defined(__aarch64__)

faiss/utils/utils.cpp

@@ -116,6 +116,8 @@ std::string get_compile_options() {
 
 #ifdef __AVX2__
     options += "AVX2 ";
+#elif __AVX512F__
+    options += "AVX512";
 #elif defined(__aarch64__)
     options += "NEON ";
 #else