Autodetect CPU architecture and AVX2/512 in cmake

CMakeLists.txt

@@ -193,45 +193,56 @@ endif ()
 if (CMAKE_BUILD_TYPE STREQUAL "")
    set(CMAKE_BUILD_TYPE "RelWithDebInfo")
 endif()
-message("CMAKE_BUILD_TYPE is ${CMAKE_BUILD_TYPE}")
+message(STATUS "CMAKE_BUILD_TYPE is ${CMAKE_BUILD_TYPE}")
 add_list_string_option(CMAKE_BUILD_TYPE "RelWithDebInfo" "Choose the type of build, options are: None(CMAKE_CXX_FLAGS or CMAKE_C_FLAGS used) Debug Release RelWithDebInfo MinSizeRel." Debug Release RelWithDebInfo MinSizeRel)
 
-Message("Architecture is ${CMAKE_SYSTEM_PROCESSOR}")
 # in case /proc/cpuinfo exists we want to inspect available Intrinsics
 # -so not to go always through SIMDE emulation
 # -so to avoid AVX512 instructions generation by gcc   
-if(EXISTS  "/proc/cpuinfo")
+execute_process(COMMAND uname -m OUTPUT_VARIABLE CPUARCH OUTPUT_STRIP_TRAILING_WHITESPACE)
-   file(STRINGS "/proc/cpuinfo" CPUINFO REGEX flags LIMIT_COUNT 1)
+message(STATUS "CPUARCH ${CPUARCH}")
-   message("AVX512 is ${AVX512}")
+if(EXISTS "/proc/cpuinfo")
-   message("AVX2 is ${AVX2}")
+  file(STRINGS "/proc/cpuinfo" CPUFLAGS REGEX flags LIMIT_COUNT 1)
-   # The following intrinsics are assumed to be available on any x86 system used to build the software (avx, f16c, fma, gnfi, mmx, pclmul, sse, sse2, sse3, xop)
+else()
-   set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX_NATIVE -DSIMDE_X86_AVX_NATIVE -DSIMDE_X86_F16C_NATIVE -DSIMDE_X86_FMA_NATIVE -DSIMDE_X86_GFNI_NATIVE -DSIMDE_X86_MMX_NATIVE -DSIMDE_X86_PCLMUL_NATIVE -DSIMDE_X86_SSE2_NATIVE -DSIMDE_X86_SSE3_NATIVE -DSIMDE_X86_SSE_NATIVE -DSIMDE_X86_XOP_HAVE_COM_ -DSIMDE_X86_XOP_NATIVE")
+  message(WARNING "did not find /proc/cpuinfo -- not setting any x86-specific compilation variables")
-   if ("${AVX512}" STREQUAL "False")
+endif()
-      set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mno-avx512f -march=native")
+
-   else()
+eval_boolean(AUTODETECT_AVX512 DEFINED CPUFLAGS AND CPUFLAGS MATCHES "avx512")
-      set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX512BW_NATIVE -DSIMDE_X86_AVX512F_NATIVE -DSIMDE_X86_AVX512VL_NATIVE -mavx512bw -march=skylake-avx512 -mtune=skylake-avx512")   
+add_boolean_option(AVX512 ${AUTODETECT_AVX512} "Whether AVX512 intrinsics is available on the host processor")
-   endif()
+
-   if (CPUINFO MATCHES "avx2" AND "${AVX2}" STREQUAL "True")
+eval_boolean(AUTODETECT_AVX2 DEFINED CPUFLAGS AND CPUFLAGS MATCHES "avx2")
-      set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX2_NATIVE -DSIMDE_X86_VPCLMULQDQ_NATIVE")
+add_boolean_option(AVX2 ${AUTODETECT_AVX2} "Whether AVX2 intrinsics is available on the host processor")
-   endif()
+
-   if (CPUINFO MATCHES "sse4_1")
+if(${CPUARCH} STREQUAL "x86_64" AND DEFINED CPUFLAGS)
-      set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_SSE4_1_NATIVE")
+  # The following intrinsics are assumed to be available on any x86 system
-   endif()
+  # (avx, f16c, fma, gnfi, mmx, pclmul, sse, sse2, sse3, xop)
-   if (CPUINFO MATCHES "sse4_2")
+  set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX_NATIVE -DSIMDE_X86_AVX_NATIVE -DSIMDE_X86_F16C_NATIVE -DSIMDE_X86_FMA_NATIVE -DSIMDE_X86_GFNI_NATIVE -DSIMDE_X86_MMX_NATIVE -DSIMDE_X86_PCLMUL_NATIVE -DSIMDE_X86_SSE2_NATIVE -DSIMDE_X86_SSE3_NATIVE -DSIMDE_X86_SSE_NATIVE -DSIMDE_X86_XOP_HAVE_COM_ -DSIMDE_X86_XOP_NATIVE")
-      set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_SSE4_2_NATIVE")
+  message(STATUS "AVX512 intrinsics are ${AVX512}")
-   endif()
+  if(${AVX512})
-   if (CPUINFO MATCHES "ssse3")
+    set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX512BW_NATIVE -DSIMDE_X86_AVX512F_NATIVE -DSIMDE_X86_AVX512VL_NATIVE -mavx512bw -march=skylake-avx512 -mtune=skylake-avx512")
-      set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_SSSE3_NATIVE")
+  else()
-   endif()
+    set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mno-avx512f -march=native")
+  endif()
+  message(STATUS "AVX2 intrinsics are ${AVX2}")
+  if(${AVX2})
+    set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX2_NATIVE -DSIMDE_X86_VPCLMULQDQ_NATIVE")
+  endif()
+  if (CPUINFO MATCHES "sse4_1")
+    set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_SSE4_1_NATIVE")
+  endif()
+  if(CPUINFO MATCHES "sse4_2")
+    set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_SSE4_2_NATIVE")
+  endif()
+  if(CPUINFO MATCHES "ssse3")
+    set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_SSSE3_NATIVE")
+  endif()
+elseif(${CPUARCH} NOT STREQUAL "x86_64")
+  message(FATAL_ERROR "Cannot compile for CPU architecture ${CPUARCH}")
 endif()
 
 set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -fno-var-tracking-assignments -march=native")
 
-Message("C_FLAGS_PROCESSOR is ${C_FLAGS_PROCESSOR}")
-
-#
 # add autotools definitions that were maybe used!
-
 add_definitions("-DSTDC_HEADERS=1 -DHAVE_SYS_TYPES_H=1 -DHAVE_SYS_STAT_H=1 -DHAVE_STDLIB_H=1 -DHAVE_STRING_H=1 -DHAVE_MEMORY_H=1 -DHAVE_STRINGS_H=1 -DHAVE_INTTYPES_H=1 -DHAVE_STDINT_H=1 -DHAVE_UNISTD_H=1 -DHAVE_FCNTL_H=1 -DHAVE_ARPA_INET_H=1 -DHAVE_SYS_TIME_H=1 -DHAVE_SYS_SOCKET_H=1 -DHAVE_STRERROR=1 -DHAVE_SOCKET=1 -DHAVE_MEMSET=1 -DHAVE_GETTIMEOFDAY=1 -DHAVE_STDLIB_H=1 -DHAVE_MALLOC=1 -DHAVE_LIBSCTP")
 
 set(commonOpts "-pipe -Wno-packed-bitfield-compat -fPIC -Wall -fno-strict-aliasing -rdynamic")

cmake_targets/build_oai

@@ -53,8 +53,6 @@ BUILD_COVERITY_SCAN=0
 DISABLE_HARDWARE_DEPENDENCY="False"
 CMAKE_BUILD_TYPE="RelWithDebInfo"
 CMAKE_CMD="$CMAKE"
-AVX512="True"
-AVX2="True"
 BUILD_ECLIPSE=0
 NR="False"
 OPTIONAL_LIBRARIES="telnetsrv enbscope uescope nrscope"
@@ -159,7 +157,7 @@ Options:
 --noavx512
    Disable AVX512 intrinsics whatever processor capability is
 --noavx2
-   Disable AVX2 intrinsics if processor supports it or use SIMDE emulation
+   Disable AVX2 intrinsics whatever processor capability is
 -k | --skip-shared-libraries
    Skip build for shared libraries to reduce compilation time when building frequently for debugging purposes
 --ninja
@@ -394,11 +392,11 @@ function main() {
             fi
             shift 2;;		
         --noavx512)
-            AVX512="False"
+            CMAKE_CMD="$CMAKE_CMD -DAVX512=OFF"
             echo_info "Disabling AVX512 instructions"
             shift 1;;
         --noavx2)
-            AVX2="False"
+            CMAKE_CMD="$CMAKE_CMD -DAVX2=OFF"
             echo_info "Disabling AVX2 instructions"
             shift 1;;
         -k | --skip-shared-libraries)
@@ -559,23 +557,9 @@ function main() {
   if [[ ${#CMAKE_C_FLAGS[@]} > 0 ]]; then CMAKE_CMD="$CMAKE_CMD -DCMAKE_C_FLAGS=\"${CMAKE_C_FLAGS[*]}\""; fi
   if [[ ${#CMAKE_CXX_FLAGS[@]} > 0 ]]; then CMAKE_CMD="$CMAKE_CMD -DCMAKE_CXX_FLAGS=\"${CMAKE_CXX_FLAGS[*]}\""; fi
 
-  # if --noavx512 is not specified the AVX512 equals "True" by default
-  # override AVX512 to "False" if processor does not have avx512
-  # this avoids having to specify --noavx512 systematically
-  if [ -f /proc/cpuinfo ]; then
-      if grep -q avx512 /proc/cpuinfo
-      then
-        echo_info "Processor has avx512 intrinsics"
-      else
-        echo_info "Processor does not have avx512 intrinsics"
-        echo_info "Forcing AVX512 to False"
-	AVX512="False"
-      fi
-  fi
-
   # for historical reasons we build in a subdirectory cmake_targets/XYZ/build,
   # e.g., cmake_targets/ran_build/build, hence the ../../..
-  CMAKE_CMD="$CMAKE_CMD -DAVX512=\"${AVX512[*]}\" -DAVX2=\"${AVX2[*]}\" ../../.."
+  CMAKE_CMD="$CMAKE_CMD ../../.."
   echo_info "running $CMAKE_CMD"
   eval $CMAKE_CMD
   

cmake_targets/macros.cmake

@@ -96,3 +96,11 @@ macro(compile_asn1 asn1Source asn1cCmd ResultFlag)
      DEPENDS ${asn1Source}
    )
 endmacro(compile_asn1)
+
+macro(eval_boolean VARIABLE)
+  if(${ARGN})
+    set(${VARIABLE} ON)
+  else()
+    set(${VARIABLE} OFF)
+  endif()
+endmacro()