Files
blis/kernels/CMakeLists.txt
Shubham Sharma 75df1ef218 Removed -fno-tree-loop-vectorize from kernel flags
- This change in made in CMAKE build system only.
- Removed -fno-tree-loop-vectorize from global kernel flags,
  instead added it to lpgemm specific kernels only.
- If this flag is not used , then gcc tries to auto
  vectorize the code which results in usages of
  vector registers, if the auto vectorized function
  is using intrinsics then the total numbers of vector
  registers used by intrinsic and auto vectorized
  code becomes more than the registers
  available in machine which causes read and writes
  to stack, which is causing regression in lpgemm.
- If this flag is enabled globally, then the files which
  do not use any intrinsic code do not get auto
  vectorized.
- To get optimal performance for both blis and lpgemm,
  this flag is enabled for lpgemm kernels only.

Change-Id: I14e5c18cd53b058bfc9d764a8eaf825b4d0a81c4
2024-07-19 00:49:52 -04:00

148 lines
7.5 KiB
CMake

##Copyright (C) 2023 - 2024, Advanced Micro Devices, Inc. All rights reserved.##
# Writing a function that will be used to generate the required object
# libraries for the required kernels.
function(generate_kernel_targets kernel_target)
# Collect all subdirectory paths that have at least one file with suffix in KERNELS_SRC_SUFS list.
get_filepaths_with_suffixes(LOCAL_SOURCE_FILES "${CMAKE_CURRENT_SOURCE_DIR}/${kernel_target}" "${KERNELS_SRC_SUFS}")
# Choose correct sub-configurarion name for the given kernel set.
get_config_for_kernel_from_kconfig_map(LOCAL_CONFIG ${kernel_target} "${KCONFIG_MAP}")
# filter the lpgemm source files to a different array
set(LOCAL_LPEGMM_SOURCE_FILES ${LOCAL_SOURCE_FILES})
list(FILTER LOCAL_SOURCE_FILES EXCLUDE REGEX ".*/lpgemm/.*")
list(FILTER LOCAL_LPEGMM_SOURCE_FILES INCLUDE REGEX ".*/lpgemm/.*")
# Only generate the object library if there is at least one source file.
list(LENGTH LOCAL_SOURCE_FILES size)
if(size GREATER 0)
# Create an object library using the source file list above.
add_library(${kernel_target}_KERNELS
OBJECT
${LOCAL_SOURCE_FILES}
)
# Include the corresponding make_defs.cmake that holds the required compiler options.
include(${CMAKE_SOURCE_DIR}/config/${LOCAL_CONFIG}/make_defs.cmake)
# Use PRIVATE keyword for option setting since we do not want the
# properties to propagate in other targets.
# mimicing get-kernel-cflags-for
target_compile_options(${kernel_target}_KERNELS
PRIVATE
# load-var-for,CKOPTFLAGS
${CKOPTFLAGS}
# load-var-for,CKVECFLAGS
${CKVECFLAGS}
# get-noopt-cflags-for
${CDBGFLAGS}
# get-noopt-cflags-for
${CWARNFLAGS}
# get-noopt-cflags-for
${CMISCFLAGS}
# get-noopt-cflags-for
${CLANGFLAGS}
# in get-kernel-cflags-for
${COMPSIMDFLAGS}
# in get-kernel-cflags-for
${BUILD_SYMFLAGS}
)
target_compile_definitions(${kernel_target}_KERNELS
PRIVATE
# in get-noopt-cflags-for
${CPPROCFLAGS}
# in get-noopt-cflags-for
${VERS_DEF}
# in get-kernel-cflags-for
${BUILD_CPPFLAGS}
)
target_include_directories(${kernel_target}_KERNELS
BEFORE
PRIVATE
# in get-noopt-cflags-for
${CINFLAGS}
)
if(THREADING_MODEL STREQUAL "openmp")
# Equivalent to CTHREADFLAGS in get-noopt-cflags-for
target_link_libraries(${kernel_target}_KERNELS PRIVATE OpenMP::OpenMP_C)
elseif(THREADING_MODEL STREQUAL "pthreads")
# in get-noopt-cflags-for
target_compile_options(${kernel_target}_KERNELS PRIVATE ${CTHREADFLAGS})
endif()
# Equivalent to CPICFLAGS in get-noopt-cflags-for
set_target_properties(${kernel_target}_KERNELS PROPERTIES POSITION_INDEPENDENT_CODE ON)
add_dependencies(${kernel_target}_KERNELS flat-header)
# Put all those targets under object-libs-targets folder name so that they appear all together in IDE.
set_target_properties(${kernel_target}_KERNELS PROPERTIES FOLDER object-libs-targets)
endif()
# Only generate the object library if there is at least one source file.
list(LENGTH LOCAL_LPEGMM_SOURCE_FILES size_lpgemm)
if (size_lpgemm GREATER 0)
# Create an object library using the source file list above.
add_library(${kernel_target}_LPGEMM_KERNELS
OBJECT
${LOCAL_LPEGMM_SOURCE_FILES}
)
# Include the corresponding make_defs.cmake that holds the required compiler options.
include(${CMAKE_SOURCE_DIR}/config/${LOCAL_CONFIG}/make_defs.cmake)
# Use PRIVATE keyword for option setting since we do not want the
# properties to propagate in other targets.
# mimicing get-kernel-cflags-for
target_compile_options(${kernel_target}_LPGEMM_KERNELS
PRIVATE
# load-var-for,CKOPTFLAGS
${CKOPTFLAGS}
# load-var-for,CKLPOPTFLAGS
${CKLPOPTFLAGS}
# load-var-for,CKVECFLAGS
${CKVECFLAGS}
# get-noopt-cflags-for
${CDBGFLAGS}
# get-noopt-cflags-for
${CWARNFLAGS}
# get-noopt-cflags-for
${CMISCFLAGS}
# get-noopt-cflags-for
${CLANGFLAGS}
# in get-kernel-cflags-for
${COMPSIMDFLAGS}
# in get-kernel-cflags-for
${BUILD_SYMFLAGS}
)
target_compile_definitions(${kernel_target}_LPGEMM_KERNELS
PRIVATE
# in get-noopt-cflags-for
${CPPROCFLAGS}
# in get-noopt-cflags-for
${VERS_DEF}
# in get-kernel-cflags-for
${BUILD_CPPFLAGS}
)
target_include_directories(${kernel_target}_LPGEMM_KERNELS
BEFORE
PRIVATE
# in get-noopt-cflags-for
${CINFLAGS}
)
if(THREADING_MODEL STREQUAL "openmp")
# Equivalent to CTHREADFLAGS in get-noopt-cflags-for
target_link_libraries(${kernel_target}_LPGEMM_KERNELS PRIVATE OpenMP::OpenMP_C)
elseif(THREADING_MODEL STREQUAL "pthreads")
# in get-noopt-cflags-for
target_compile_options(${kernel_target}_LPGEMM_KERNELS PRIVATE ${CTHREADFLAGS})
endif()
# Equivalent to CPICFLAGS in get-noopt-cflags-for
set_target_properties(${kernel_target}_LPGEMM_KERNELS PROPERTIES POSITION_INDEPENDENT_CODE ON)
add_dependencies(${kernel_target}_LPGEMM_KERNELS flat-header)
# Put all those targets under object-libs-targets folder name so that they appear
# all together in IDE.
set_target_properties(${kernel_target}_LPGEMM_KERNELS PROPERTIES FOLDER object-libs-targets)
endif()
endfunction()
# Generate targets for each of the kernels present
# in the kernel list.
foreach(KERN ${KERNEL_LIST})
generate_kernel_targets(${KERN})
endforeach()