diff options
Diffstat (limited to 'volk/lib')
| -rw-r--r-- | volk/lib/CMakeLists.txt | 364 | ||||
| -rw-r--r-- | volk/lib/gcc_x86_cpuid.h | 6 | ||||
| -rw-r--r-- | volk/lib/testqa.cc | 2 |
3 files changed, 172 insertions, 200 deletions
diff --git a/volk/lib/CMakeLists.txt b/volk/lib/CMakeLists.txt index b491f94bb..8288786c9 100644 --- a/volk/lib/CMakeLists.txt +++ b/volk/lib/CMakeLists.txt @@ -1,5 +1,5 @@ # -# Copyright 2011 Free Software Foundation, Inc. +# Copyright 2011-2012 Free Software Foundation, Inc. # # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by @@ -16,6 +16,31 @@ # ######################################################################## +# header file detection +######################################################################## +include(CheckIncludeFile) +CHECK_INCLUDE_FILE(cpuid.h HAVE_CPUID_H) +if(HAVE_CPUID_H) + add_definitions(-DHAVE_CPUID_H) +endif() + +CHECK_INCLUDE_FILE(intrin.h HAVE_INTRIN_H) +if(HAVE_INTRIN_H) + add_definitions(-DHAVE_INTRIN_H) +endif() + +CHECK_INCLUDE_FILE(fenv.h HAVE_FENV_H) +if(HAVE_FENV_H) + add_definitions(-DHAVE_FENV_H) +endif() + +CHECK_INCLUDE_FILE(dlfcn.h HAVE_DLFCN_H) +if(HAVE_DLFCN_H) + add_definitions(-DHAVE_DLFCN_H) + list(APPEND volk_libraries ${CMAKE_DL_LIBS}) +endif() + +######################################################################## # Setup the compiler name ######################################################################## set(COMPILER_NAME ${CMAKE_C_COMPILER_ID}) @@ -23,235 +48,183 @@ if(MSVC) #its not set otherwise set(COMPILER_NAME MSVC) endif() +message(STATUS "Compiler name: ${COMPILER_NAME}") + if(NOT DEFINED COMPILER_NAME) message(FATAL_ERROR "COMPILER_NAME undefined. Volk build may not support this compiler.") endif() ######################################################################## -# Parse the arches xml file: -# Test each arch to see if the compiler supports the flag. -# If the test passes append the arch to the available list. +# detect x86 flavor of CPU ######################################################################## -#extract the compiler lines from the xml file using abusive python - - - -execute_process( - COMMAND ${PYTHON_EXECUTABLE} -c - "from xml.dom import minidom; print ';'.join(map(lambda b: ','.join([','.join([b.attributes['name'].value,item.attributes['name'].value,item.firstChild.data]) for item in b.getElementsByTagName('remap')]), minidom.parse('${CMAKE_SOURCE_DIR}/gen/compilers.xml').getElementsByTagName('compiler')))" - - OUTPUT_VARIABLE compiler_lines OUTPUT_STRIP_TRAILING_WHITESPACE -) - -foreach(thing ${compiler_lines}) - string(REGEX REPLACE "," ";" thing_list ${thing}) - list(FIND thing_list ${COMPILER_NAME} check_val) - if(NOT ("${check_val}" STREQUAL "-1")) - string(REGEX REPLACE "${COMPILER_NAME}," ";" filter_string ${thing}) - endif() -endforeach() - +if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "^(i.86|x86|x86_64|amd64)$") + message(STATUS "x86* CPU detected") + set(CPU_IS_x86 TRUE) +endif() -#extract compiler prefixes from the xml file using abusive python +######################################################################## +# determine passing architectures based on compile flag tests +######################################################################## execute_process( - COMMAND ${PYTHON_EXECUTABLE} -c - "from xml.dom import minidom; print ';'.join(map(lambda b: ','.join([','.join([b.attributes['name'].value,item.firstChild.data]) for item in b.getElementsByTagName('prefix')]), minidom.parse('${CMAKE_SOURCE_DIR}/gen/compilers.xml').getElementsByTagName('compiler')))" - - OUTPUT_VARIABLE compiler_prefixes OUTPUT_STRIP_TRAILING_WHITESPACE + COMMAND ${PYTHON_EXECUTABLE} ${PYTHON_DASH_B} + ${CMAKE_SOURCE_DIR}/gen/volk_compile_utils.py + --mode "arch_flags" --compiler "${COMPILER_NAME}" + OUTPUT_VARIABLE arch_flag_lines OUTPUT_STRIP_TRAILING_WHITESPACE ) -foreach(thing ${compiler_prefixes}) - string(REGEX REPLACE "," ";" thing_list ${thing}) - list(FIND thing_list ${COMPILER_NAME} check_val) - if(NOT ("${check_val}" STREQUAL "-1")) - list(GET thing_list "1" prefix) +macro(check_arch arch_name) + set(flags ${ARGN}) + set(have_${arch_name} TRUE) + foreach(flag ${flags}) + include(CheckCXXCompilerFlag) + set(have_flag have${flag}) + execute_process( #make the have_flag have nice alphanum chars (just for looks/not necessary) + COMMAND ${PYTHON_EXECUTABLE} -c "import re; print(re.sub('\\W', '_', '${have_flag}'))" + OUTPUT_VARIABLE have_flag OUTPUT_STRIP_TRAILING_WHITESPACE + ) + CHECK_CXX_COMPILER_FLAG(${flag} ${have_flag}) + if (NOT ${have_flag}) + set(have_${arch_name} FALSE) + endif() + endforeach(flag) + if (have_${arch_name}) + list(APPEND available_archs ${arch_name}) endif() -endforeach() - - - - -#extract the arch lines from the xml file using abusive python -execute_process( - COMMAND ${PYTHON_EXECUTABLE} -c - "from xml.dom import minidom; print ';'.join(map(lambda a: '%s %s %s %s'%(a.attributes['name'].value,a.getElementsByTagName('flag')[0].firstChild.data,a.getElementsByTagName('overrule')[0].firstChild.data,a.getElementsByTagName('overrule_val')[0].firstChild.data) if (len(a.getElementsByTagName('overrule'))) else '%s %s %s %s'%(a.attributes['name'].value,a.getElementsByTagName('flag')[0].firstChild.data,'no_overrule', 'no_overrule_val'), minidom.parse('${CMAKE_SOURCE_DIR}/gen/archs.xml').getElementsByTagName('arch')))" - - OUTPUT_VARIABLE arch_lines OUTPUT_STRIP_TRAILING_WHITESPACE -) +endmacro(check_arch) +foreach(line ${arch_flag_lines}) + string(REGEX REPLACE "," ";" arch_flags ${line}) + check_arch(${arch_flags}) +endforeach(line) +macro(OVERRULE_ARCH arch reason) + message(STATUS "${reason}, Overruled arch ${arch}") + list(REMOVE_ITEM available_archs ${arch}) +endmacro(OVERRULE_ARCH) - -#set the various overrule values (see archs.xml) -#a lot of this is translating between automake and cmake -if(NOT "${CROSSCOMPILE_MULTILIB}" STREQUAL "true") - set(MD_SUBCPU ${CMAKE_SYSTEM_PROCESSOR}) - #detect 32 or 64 bit compiler - if(MD_SUBCPU MATCHES "^(i.86|x86|x86_64|amd64)$") - include(CheckTypeSize) - check_type_size("void*" SIZEOF_VOID_P BUILTIN_TYPES_ONLY) - if (${SIZEOF_VOID_P} EQUAL 8) - set(MD_SUBCPU x86_64) - else() - set(MD_SUBCPU x86) - endif() +######################################################################## +# eliminate AVX on GCC < 4.4 +# even though it accepts -mavx, as won't assemble xgetbv, which we need +######################################################################## +if(CPU_IS_x86 AND COMPILER_NAME MATCHES "GNU") + execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion + OUTPUT_VARIABLE GCC_VERSION OUTPUT_STRIP_TRAILING_WHITESPACE) + if(GCC_VERSION VERSION_LESS "4.4") + OVERRULE_ARCH(avx "GCC missing xgetbv") endif() endif() -if(NOT "${ORC_FOUND}" STREQUAL "TRUE") - set(LV_HAVE_ORC "no") -endif() - +######################################################################## +# implement overruling in the ORC case, +# since ORC always passes flag detection +######################################################################## +if(NOT ORC_FOUND) + OVERRULE_ARCH(orc "ORC support not found") +endif() - - - -macro(compiler_filter name flag) - set(filtered_flag ${flag}) - foreach(thing ${filter_string}) - string(REGEX REPLACE "," ";" flagmap ${thing}) - list(GET flagmap "0" key) - list(GET flagmap "1" val) - string(REGEX MATCH "^${key}$" found ${flag}) - if("${found}" STREQUAL "${key}") - string(REGEX REPLACE "^${key}$" "${val}" filtered_flag ${flag}) - endif() - endforeach() - set(${name}_flag "${prefix}${filtered_flag}") -endmacro() - - - - - - - -macro(handle_arch name flag overrule overrule_val) - - #handle overrule - if("${${overrule}}" STREQUAL "${overrule_val}") - set(have_${name} FALSE) - message(STATUS "${name} overruled") - #handle special case for none flag - elseif(${flag} STREQUAL "none") - set(have_${name} TRUE) - #otherwise test the flag(s) against the compiler - else() - include(CheckCXXCompilerFlag) - string(REGEX REPLACE "," ";" flag_list ${flag}) - set(have_${name} 1) - foreach(thing ${flag_list}) - compiler_filter(${name} ${thing}) - CHECK_CXX_COMPILER_FLAG(${${name}_flag} have_${thing}) - if(NOT (${have_${name}} AND ("${have_${thing}}" STREQUAL "1"))) - set(have_${name} 0) - endif() - endforeach() +######################################################################## +# implement overruling in the non-multilib case +# this makes things work when both -m32 and -m64 pass +######################################################################## +if(NOT CROSSCOMPILE_MULTILIB AND CPU_IS_x86) + include(CheckTypeSize) + check_type_size("void*[8]" SIZEOF_CPU BUILTIN_TYPES_ONLY) + if (${SIZEOF_CPU} EQUAL 64) + OVERRULE_ARCH(32 "CPU width is 64 bits") endif() - - if(have_${name}) - list(APPEND available_arches ${name}) + if (${SIZEOF_CPU} EQUAL 32) + OVERRULE_ARCH(64 "CPU width is 32 bits") endif() +endif() -endmacro(handle_arch) - -#create a list of available arches -foreach(arch_line ${arch_lines}) - string(REPLACE " " ";" args "${arch_line}") - handle_arch(${args}) -endforeach(arch_line) - -message(STATUS "Available arches: ${available_arches}") +######################################################################## +# done overrules! print the result +######################################################################## +message(STATUS "Available architectures: ${available_archs}") ######################################################################## -# Parse the machines xml file: -# Test each machine to see if its arch dependencies are supported. -# Build a list of supported machines and the machine definitions. +# determine available machines given the available architectures ######################################################################## -#extract the machine lines from the xml file using crazy python execute_process( - COMMAND ${PYTHON_EXECUTABLE} -c - "from xml.dom import minidom; print ';'.join(map(lambda a: '%s %s'%(a.attributes['name'].value,a.getElementsByTagName('archs')[0].firstChild.data),minidom.parse('${CMAKE_SOURCE_DIR}/gen/machines.xml').getElementsByTagName('machine')))" - OUTPUT_VARIABLE machine_lines OUTPUT_STRIP_TRAILING_WHITESPACE + COMMAND ${PYTHON_EXECUTABLE} ${PYTHON_DASH_B} + ${CMAKE_SOURCE_DIR}/gen/volk_compile_utils.py + --mode "machines" --archs "${available_archs}" + OUTPUT_VARIABLE available_machines OUTPUT_STRIP_TRAILING_WHITESPACE ) -macro(handle_machine1 name) - unset(machine_flags) - string(TOUPPER LV_MACHINE_${name} machine_def) - - #check if all the arches are supported - foreach(arch ${ARGN}) - set(is_match ${have_${arch}}) - if(NOT is_match) - set(is_match FALSE) - break() - endif(NOT is_match) - set(machine_flags "${machine_flags} ${${arch}_flag}") - endforeach(arch) - - string(REGEX REPLACE "^[ \t]+" "" machine_flags "${machine_flags}") - - if(is_match) - #this is a match, append the source and set its flags - set(machine_source ${CMAKE_CURRENT_BINARY_DIR}/volk_machine_${name}.c) - set_source_files_properties(${machine_source} PROPERTIES COMPILE_FLAGS "${machine_flags}") - list(APPEND machine_sources ${machine_source}) - list(APPEND machine_defs ${machine_def}) - list(APPEND available_machines ${name}) - endif() -endmacro(handle_machine1) - -macro(handle_machine name) - set(arches ${ARGN}) - list(FIND arches "32|64" index) - if(${index} EQUAL -1) - handle_machine1(${name} ${arches}) - else() - list(REMOVE_ITEM arches "32|64") - handle_machine1(${name}_32 32 ${arches}) - handle_machine1(${name}_64 64 ${arches}) - endif() -endmacro(handle_machine) - -#setup the available machines -foreach(machine_line ${machine_lines}) - string(REPLACE " " ";" args "${machine_line}") - handle_machine(${args}) -endforeach(machine_line) +######################################################################## +# Implement machine overruling for redundant machines: +# A machine is redundant when expansion rules occur, +# and the arch superset passes configuration checks. +# When this occurs, eliminate the redundant machines +# to avoid unnecessary compilation of subset machines. +######################################################################## +foreach(arch orc 64 32) + foreach(machine_name ${available_machines}) + string(REPLACE "_${arch}" "" machine_name_no_arch ${machine_name}) + if (${machine_name} STREQUAL ${machine_name_no_arch}) + else() + list(REMOVE_ITEM available_machines ${machine_name_no_arch}) + endif() + endforeach(machine_name) +endforeach(arch) +######################################################################## +# done overrules! print the result +######################################################################## message(STATUS "Available machines: ${available_machines}") ######################################################################## # Create rules to run the volk generator ######################################################################## -#list of the generated sources -set(volk_gen_sources - ${CMAKE_BINARY_DIR}/include/volk/volk.h - ${CMAKE_BINARY_DIR}/lib/volk.c - ${CMAKE_BINARY_DIR}/lib/volk_init.h - ${CMAKE_BINARY_DIR}/include/volk/volk_typedefs.h - ${CMAKE_BINARY_DIR}/include/volk/volk_cpu.h - ${CMAKE_BINARY_DIR}/lib/volk_cpu.c - ${CMAKE_BINARY_DIR}/include/volk/volk_config_fixed.h - ${CMAKE_BINARY_DIR}/lib/volk_environment_init.c - ${CMAKE_BINARY_DIR}/lib/volk_environment_init.h - ${CMAKE_BINARY_DIR}/lib/volk_machines.h - ${CMAKE_BINARY_DIR}/lib/volk_machines.c - ${machine_sources} -) #dependencies are all python, xml, and header implementation files file(GLOB xml_files ${CMAKE_SOURCE_DIR}/gen/*.xml) file(GLOB py_files ${CMAKE_SOURCE_DIR}/gen/*.py) file(GLOB h_files ${CMAKE_SOURCE_DIR}/include/volk/*.h) -add_custom_command( - OUTPUT ${volk_gen_sources} - DEPENDS ${xml_files} ${py_files} ${h_files} - COMMAND ${PYTHON_EXECUTABLE} -B - ${CMAKE_SOURCE_DIR}/gen/volk_register.py - ${CMAKE_BINARY_DIR} -) +macro(gen_template tmpl output) + list(APPEND volk_gen_sources ${output}) + add_custom_command( + OUTPUT ${output} + DEPENDS ${xml_files} ${py_files} ${h_files} ${tmpl} + COMMAND ${PYTHON_EXECUTABLE} ${PYTHON_DASH_B} + ${CMAKE_SOURCE_DIR}/gen/volk_tmpl_utils.py + --input ${tmpl} --output ${output} ${ARGN} + ) +endmacro(gen_template) + +make_directory(${CMAKE_BINARY_DIR}/include/volk) + +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk.tmpl.h ${CMAKE_BINARY_DIR}/include/volk/volk.h) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk.tmpl.c ${CMAKE_BINARY_DIR}/lib/volk.c) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_typedefs.tmpl.h ${CMAKE_BINARY_DIR}/include/volk/volk_typedefs.h) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_cpu.tmpl.h ${CMAKE_BINARY_DIR}/include/volk/volk_cpu.h) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_cpu.tmpl.c ${CMAKE_BINARY_DIR}/lib/volk_cpu.c) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_config_fixed.tmpl.h ${CMAKE_BINARY_DIR}/include/volk/volk_config_fixed.h) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_machines.tmpl.h ${CMAKE_BINARY_DIR}/lib/volk_machines.h) +gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_machines.tmpl.c ${CMAKE_BINARY_DIR}/lib/volk_machines.c) + +foreach(machine_name ${available_machines}) + #generate machine source + set(machine_source ${CMAKE_CURRENT_BINARY_DIR}/volk_machine_${machine_name}.c) + gen_template(${CMAKE_SOURCE_DIR}/tmpl/volk_machine_xxx.tmpl.c ${machine_source} ${machine_name}) + + #determine machine flags + execute_process( + COMMAND ${PYTHON_EXECUTABLE} ${PYTHON_DASH_B} + ${CMAKE_SOURCE_DIR}/gen/volk_compile_utils.py + --mode "machine_flags" --machine "${machine_name}" --compiler "${COMPILER_NAME}" + OUTPUT_VARIABLE ${machine_name}_flags OUTPUT_STRIP_TRAILING_WHITESPACE + ) + if(${machine_name}_flags) + set_source_files_properties(${machine_source} PROPERTIES COMPILE_FLAGS "${${machine_name}_flags}") + endif() + + #add to available machine defs + string(TOUPPER LV_MACHINE_${machine_name} machine_def) + list(APPEND machine_defs ${machine_def}) +endforeach(machine_name) ######################################################################## # Set local include directories first @@ -270,7 +243,7 @@ if(ORC_FOUND) #setup orc library usage include_directories(${ORC_INCLUDE_DIRS}) link_directories(${ORC_LIBRARY_DIRS}) - add_definitions(-DLV_HAVE_ORC) + list(APPEND volk_libraries ${ORC_LIBRARIES}) #setup orc functions file(GLOB orc_files ${CMAKE_SOURCE_DIR}/orc/*.orc) @@ -313,16 +286,15 @@ PROPERTIES COMPILE_DEFINITIONS "${machine_defs}") if(MSVC) #add compatibility includes for stdint types - include_directories(${CMAKE_SOURCE_DIR}/msvc) + include_directories(${CMAKE_SOURCE_DIR}/cmake/msvc) + add_definitions(-DHAVE_CONFIG_H) #compile the sources as C++ due to the lack of complex.h under MSVC set_source_files_properties(${volk_sources} PROPERTIES LANGUAGE CXX) endif() #create the volk runtime library add_library(volk SHARED ${volk_sources}) -if(ORC_FOUND) - target_link_libraries(volk ${ORC_LIBRARIES}) -endif(ORC_FOUND) +target_link_libraries(volk ${volk_libraries}) set_target_properties(volk PROPERTIES SOVERSION ${LIBVER}) set_target_properties(volk PROPERTIES DEFINE_SYMBOL "volk_EXPORTS") diff --git a/volk/lib/gcc_x86_cpuid.h b/volk/lib/gcc_x86_cpuid.h index e0254f192..3c3f47b00 100644 --- a/volk/lib/gcc_x86_cpuid.h +++ b/volk/lib/gcc_x86_cpuid.h @@ -5,16 +5,16 @@ * under the terms of the GNU General Public License as published by the * Free Software Foundation; either version 3, or (at your option) any * later version. - * + * * This file is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. - * + * * Under Section 7 of GPL version 3, you are granted additional * permissions described in the GCC Runtime Library Exception, version * 3.1, as published by the Free Software Foundation. - * + * * You should have received a copy of the GNU General Public License and * a copy of the GCC Runtime Library Exception along with this program; * see the files COPYING3 and COPYING.RUNTIME respectively. If not, see diff --git a/volk/lib/testqa.cc b/volk/lib/testqa.cc index 593087f85..d0204fc01 100644 --- a/volk/lib/testqa.cc +++ b/volk/lib/testqa.cc @@ -55,7 +55,7 @@ VOLK_RUN_TESTS(volk_32f_x2_dot_prod_32f_a, 1e-4, 0, 204600, 1); VOLK_RUN_TESTS(volk_32f_x2_dot_prod_32f_u, 1e-4, 0, 204600, 1); //VOLK_RUN_TESTS(volk_32f_s32f_32f_fm_detect_32f_a, 1e-4, 2046, 10000); VOLK_RUN_TESTS(volk_32f_index_max_16u_a, 3, 0, 20460, 1); -VOLK_RUN_TESTS(volk_32f_x2_s32f_interleave_16ic_a, 1, 32768, 20460, 1); +VOLK_RUN_TESTS(volk_32f_x2_s32f_interleave_16ic_a, 1, 32767, 20460, 1); VOLK_RUN_TESTS(volk_32f_x2_interleave_32fc_a, 0, 0, 20460, 1); VOLK_RUN_TESTS(volk_32f_x2_max_32f_a, 1e-4, 0, 20460, 1); VOLK_RUN_TESTS(volk_32f_x2_min_32f_a, 1e-4, 0, 20460, 1); |