Add Intrinsics usage by SIMDE

This commit is contained in:
Bruno Mongazon-Cazavet
2022-09-28 18:21:19 +02:00
committed by Robert Schmidt
parent a1de5e3d11
commit fe1dc24e28
114 changed files with 7900 additions and 11031 deletions
+33 -53
View File
@@ -178,7 +178,6 @@ if (CUDA_FOUND)
# Disable warnings for CUDA
SET(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};-lpthread;-w;-O3;--default-stream;per-thread;-I/usr/local/cuda/inc;-L/usr/local/cuda/lib -lcutil;-rdc=true;-lcudadevrt")
SET(CUDA_VERBOSE_BUILD ON)
SET(CUDA_HOST_COMPILER "/usr/bin/g++")
SET(CUDA_SEPARABLE_COMPILATION ON)
@@ -198,44 +197,36 @@ message("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}")
if (CMAKE_SYSTEM_PROCESSOR STREQUAL "armv7l")
set(C_FLAGS_PROCESSOR "-gdwarf-2 -mfloat-abi=hard -mfpu=neon -lgcc -lrt")
else (CMAKE_SYSTEM_PROCESSOR STREQUAL "armv7l")
if(EXISTS "/proc/cpuinfo")
file(STRINGS "/proc/cpuinfo" CPUINFO REGEX flags LIMIT_COUNT 1)
message("NOAVX512 is ${NOAVX512}")
if (CPUINFO MATCHES "avx512bw" AND "${NOAVX512}" STREQUAL "False")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mavx512bw -march=skylake-avx512 -mtune=skylake-avx512 " )
set(COMPILATION_AVX2 "True")
else()
if (CPUINFO MATCHES "avx2")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mavx2")
set(COMPILATION_AVX2 "True")
else()
set(COMPILATION_AVX2 "False")
endif()
if (CPUINFO MATCHES "sse4_1")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -msse4.1 -mpclmul")
endif()
if (CPUINFO MATCHES "ssse3")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mssse3")
endif()
endif()
else()
Message("/proc/cpuinfo does not exit. We will use manual CPU flags")
endif()
# 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")
file(STRINGS "/proc/cpuinfo" CPUINFO REGEX flags LIMIT_COUNT 1)
message("AVX512 is ${AVX512}")
message("AVX2 is ${AVX2}")
if ("${AVX512}" STREQUAL "False")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mno-avx512f")
else()
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -mavx512bw -march=skylake-avx512 -mtune=skylake-avx512")
endif()
if (CPUINFO MATCHES "avx2" AND "${AVX2}" STREQUAL "True")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -DSIMDE_X86_AVX2_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()
endif()
set(C_FLAGS_PROCESSOR " ${C_FLAGS_PROCESSOR} ${CFLAGS_PROCESSOR_USER}")
set(C_FLAGS_PROCESSOR "${C_FLAGS_PROCESSOR} -fno-var-tracking-assignments -march=native")
Message("C_FLAGS_PROCESSOR is ${C_FLAGS_PROCESSOR}")
#if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86")
# if ( (NOT( C_FLAGS_PROCESSOR MATCHES "ssse3")) OR (NOT( C_FLAGS_PROCESSOR MATCHES "msse4.1")) )
# Message(FATAL_ERROR "For x86 Architecture, you must have following flags: -mssse3 -msse4.1. The current detected flags are: ${C_FLAGS_PROCESSOR}. You can pass the flags manually in build script, for example: ./build_oai --cflags_processor \"-mssse3 -msse4.1 -mavx2\" ")
# endif()
#endif()
#
# add autotools definitions that were maybe used!
@@ -855,9 +846,6 @@ endif ()
include_directories ("${OPENAIR_DIR}/sdr/COMMON")
Message("DEADLINE_SCHEDULER flag is ${DEADLINE_SCHEDULER}")
Message("CPU_Affinity flag is ${CPU_AFFINITY}")
##############################################################
# ???!!! TO BE DOCUMENTED OPTIONS !!!???
##############################################################
@@ -1711,14 +1699,8 @@ if (${SMBV})
set(PHY_SRC "${PHY_SRC} ${OPENAIR1_DIR}/PHY/TOOLS/smbv.c")
endif (${SMBV})
if (${COMPILATION_AVX2} STREQUAL "True")
#set(PHY_SRC ${PHY_SRC} ${OPENAIR1_DIR}/PHY/LTE_TRANSPORT/dlsch_llr_computation_avx2.c)
set(PHY_SRC_UE ${PHY_SRC_UE} ${OPENAIR1_DIR}/PHY/LTE_UE_TRANSPORT/dlsch_llr_computation_avx2.c)
endif ()
if (${COMPILATION_AVX2} STREQUAL "True")
set(PHY_NR_UE_SRC ${PHY_NR_UE_SRC} ${OPENAIR1_DIR}/PHY/LTE_UE_TRANSPORT/dlsch_llr_computation_avx2.c)
endif ()
set(PHY_SRC_UE ${PHY_SRC_UE} ${OPENAIR1_DIR}/PHY/LTE_UE_TRANSPORT/dlsch_llr_computation_avx2.c)
set(PHY_NR_UE_SRC ${PHY_NR_UE_SRC} ${OPENAIR1_DIR}/PHY/LTE_UE_TRANSPORT/dlsch_llr_computation_avx2.c)
add_library(PHY_COMMON ${PHY_SRC_COMMON})
add_dependencies(PHY_COMMON rrc_flag)
@@ -2829,16 +2811,14 @@ add_executable(nr-uesoftmodem
target_link_libraries (nr-uesoftmodem
-Wl,--start-group
RRC_LIB NR_RRC_LIB NGAP_LIB NGAP_GNB SECU_CN SECU_OSA UTIL HASHTABLE SCTP_CLIENT SCHED_RU_LIB SCHED_UE_LIB SCHED_NR_UE_LIB
PHY_COMMON PHY_NR_COMMON PHY_UE PHY_NR_UE PHY_RU NR_L2_UE L2_UE_LTE_NR MAC_NR_COMMON NFAPI_COMMON_LIB NFAPI_LIB NFAPI_PNF_LIB
NFAPI_USER_LIB MISC_NFAPI_NR_LIB S1AP_LIB S1AP_ENB
${RAL_LIB} ${NAS_UE_LIB} ITTI ${FLPT_MSG_LIB} ${ATLAS_LIBRARIES}
NFAPI_USER_LIB S1AP_LIB S1AP_ENB
${RAL_LIB} ${NAS_UE_LIB} ITTI ${FLPT_MSG_LIB} ${ATLAS_LIBRARIES} LIB_5GNAS_GNB LIB_NAS_SIMUE ${NAS_SIM_LIB}
NR_RRC_LIB SECU_CN SECU_OSA UTIL HASHTABLE SCHED_RU_LIB SCHED_NR_UE_LIB
PHY_COMMON PHY_NR_COMMON PHY_NR_UE NR_L2_UE L2_UE_LTE_NR MAC_NR_COMMON NFAPI_COMMON_LIB NFAPI_LIB NFAPI_PNF_LIB
NFAPI_USER_LIB MISC_NFAPI_NR_LIB
${RAL_LIB} ITTI ${ATLAS_LIBRARIES} LIB_5GNAS_GNB LIB_NAS_SIMUE ${NAS_SIM_LIB}
-Wl,--end-group z dl)
target_link_libraries (nr-uesoftmodem ${LIBXML2_LIBRARIES})
target_link_libraries (nr-uesoftmodem pthread m CONFIG_LIB rt crypt ${CRYPTO_LIBRARIES} ${OPENSSL_LIBRARIES} sctp ${XFORMS_LIBRARIES} ${CMAKE_DL_LIBS} ${LIBYAML_LIBRARIES} ${ATLAS_LIBRARIES})
target_link_libraries (nr-uesoftmodem pthread m CONFIG_LIB rt crypt ${CRYPTO_LIBRARIES} ${OPENSSL_LIBRARIES} ${XFORMS_LIBRARIES} ${CMAKE_DL_LIBS} ${LIBYAML_LIBRARIES} ${ATLAS_LIBRARIES})
target_link_libraries (nr-uesoftmodem ${LIB_LMS_LIBRARIES})
target_link_libraries (nr-uesoftmodem ${T_LIB})
@@ -3126,7 +3106,7 @@ function(make_driver name dir)
endforeach()
CONFIGURE_FILE(${OPENAIR_CMAKE}/tools/Kbuild.cmake ${OPENAIR_BIN_DIR}/${name}/Kbuild)
add_custom_command(OUTPUT ${name}.ko
COMMAND make -j2 -C ${module_build_path} M=${OPENAIR_BIN_DIR}/${name}
COMMAND make -C ${module_build_path} M=${OPENAIR_BIN_DIR}/${name}
WORKING_DIRECTORY ${OPENAIR_BIN_DIR}/${name}
COMMENT "building ${module}.ko"
VERBATIM