Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

This patch adds more options for unrolling DFT kernels #659

Merged
merged 1 commit into from
Mar 15, 2025
Merged

This patch adds more options for unrolling DFT kernels #659

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
82 changes: 71 additions & 11 deletions src/dft/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,12 +11,29 @@ if (SLEEFDFT_MAXBUTWIDTH GREATER 7)
message(FATAL_ERROR "SLEEFDFT_MAXBUTWIDTH has to be smaller than 8." )
endif()

set(SLEEFDFT_MAXSHIFT 2 CACHE STRING "Max hardcoded shift")
set(SLEEFDFT_MINSHIFT 1 CACHE STRING "Min hardcoded shift")
set(SLEEFDFT_MAXSHIFT 1 CACHE STRING "Max hardcoded shift")

if ((${SLEEFDFT_MINSHIFT} GREATER ${SLEEFDFT_MAXSHIFT}) OR (${SLEEFDFT_MINSHIFT} LESS 1))
message(FATAL_ERROR "SLEEFDFT_MINSHIFT, SLEEFDFT_MAXSHIFT range error")
endif()

math(EXPR SLEEFDFT_MAXSHIFT_MINUS_1 "${SLEEFDFT_MAXSHIFT} - 1")
foreach(J RANGE 0 ${SLEEFDFT_MAXSHIFT_MINUS_1})
list(APPEND LISTSHIFTSTR ${J})
endforeach()
if (${SLEEFDFT_MINSHIFT} LESS ${SLEEFDFT_MAXSHIFT})
foreach(J RANGE ${SLEEFDFT_MINSHIFT} ${SLEEFDFT_MAXSHIFT_MINUS_1})
list(APPEND LISTSHIFTSTR ${J})
endforeach()
else()
set(LISTSHIFTSTR)
endif()

if (1 LESS ${SLEEFDFT_MAXSHIFT})
foreach(J RANGE 1 ${SLEEFDFT_MAXSHIFT_MINUS_1})
list(APPEND LISTSHIFTSTR2 ${J})
endforeach()
else()
set(LISTSHIFTSTR2)
endif()

# Settings

Expand All @@ -30,6 +47,10 @@ set(MACRODEF_purecdp BASETYPEID=1 ENABLE_PUREC CONFIG=1)
set(CFLAGS_purecdp ${FLAGS_ENABLE_PUREC})
set(MACRODEF_purecsp BASETYPEID=2 ENABLE_PUREC CONFIG=1)
set(CFLAGS_purecsp ${FLAGS_ENABLE_PUREC})
if(CMAKE_C_COMPILER_ID MATCHES "Clang")
set(CFLAGS_purecdp ${FLAGS_ENABLE_PUREC} -O0)
set(CFLAGS_purecsp ${FLAGS_ENABLE_PUREC} -O0)
endif()
set(MACRODEF_purecld BASETYPEID=3 ENABLE_PUREC CONFIG=1)
set(CFLAGS_purecld ${FLAGS_ENABLE_PUREC})
set(MACRODEF_purecqp BASETYPEID=4 ENABLE_PUREC CONFIG=1)
Expand Down Expand Up @@ -215,7 +236,8 @@ endif()

set(COMMON_TARGET_DEFINITIONS ${COMMON_TARGET_DEFINITIONS}
MAXBUTWIDTHDP=${SLEEFDFT_MAXBUTWIDTH} MAXBUTWIDTHSP=${SLEEFDFT_MAXBUTWIDTH}
MAXSHIFTDP=${SLEEFDFT_MAXSHIFT} MAXSHIFTSP=${SLEEFDFT_MAXSHIFT}
MINSHIFTDP=${SLEEFDFT_MINSHIFT} MAXSHIFTDP=${SLEEFDFT_MAXSHIFT}
MINSHIFTSP=${SLEEFDFT_MINSHIFT} MAXSHIFTSP=${SLEEFDFT_MAXSHIFT}
)

if (SLEEFDFT_ENABLE_STREAM)
Expand Down Expand Up @@ -253,7 +275,7 @@ endif()

add_custom_command(OUTPUT dispatchparam.h
COMMENT "Generating dispatchparam.h"
COMMAND $<TARGET_FILE:${TARGET_MKDISPATCH}> paramonly ALL ${SLEEFDFT_MAXBUTWIDTH} ${SLEEFDFT_MAXSHIFT} ${ISALIST_SP} > ${CMAKE_CURRENT_BINARY_DIR}/dispatchparam.h
COMMAND $<TARGET_FILE:${TARGET_MKDISPATCH}> paramonly ALL ${SLEEFDFT_MAXBUTWIDTH} ${SLEEFDFT_MINSHIFT} ${SLEEFDFT_MAXSHIFT} ${ISALIST_SP} > ${CMAKE_CURRENT_BINARY_DIR}/dispatchparam.h
DEPENDS ${TARGET_MKDISPATCH}
)
add_custom_target(dispatchparam.h_generated SOURCES ${CMAKE_CURRENT_BINARY_DIR}/dispatchparam.h)
Expand All @@ -269,7 +291,7 @@ foreach(T ${LIST_SUPPORTED_FPTYPE})
string(CONCAT S "dispatch" ${ST} ".hpp") # S is dispatchdp.hpp
add_custom_command(OUTPUT ${S}
COMMENT "Generating ${S}"
COMMAND $<TARGET_FILE:${TARGET_MKDISPATCH}> ${LT} ${CST} ${SLEEFDFT_MAXBUTWIDTH} ${SLEEFDFT_MAXSHIFT} ${ISALIST_${CST}} > ${S}
COMMAND $<TARGET_FILE:${TARGET_MKDISPATCH}> ${LT} ${CST} ${SLEEFDFT_MAXBUTWIDTH} ${SLEEFDFT_MINSHIFT} ${SLEEFDFT_MAXSHIFT} ${ISALIST_${CST}} > ${S}
DEPENDS ${TARGET_MKDISPATCH}
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
)
Expand Down Expand Up @@ -305,6 +327,11 @@ add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/unroll1.cpp.in
DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/unroll1.cpp.in)
add_custom_target(unroll1.cpp.in.copied DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/unroll1.cpp.in)

add_custom_command(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/unroll2.cpp.in
COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/unroll2.cpp.in ${CMAKE_CURRENT_BINARY_DIR}
DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/unroll2.cpp.in)
add_custom_target(unroll2.cpp.in.copied DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/unroll2.cpp.in)

# Target unroll*.cpp

foreach(T ${LIST_SUPPORTED_FPTYPE})
Expand All @@ -323,7 +350,7 @@ foreach(T ${LIST_SUPPORTED_FPTYPE})
if(UNROLL_TARGET_${CST})
add_custom_command(OUTPUT ${UNROLL_TARGET_${CST}}
COMMENT "Generating ${UNROLL_TARGET_${CST}}"
COMMAND $<TARGET_FILE:${TARGET_MKUNROLL}> unroll0.cpp.in ${LT} ${CST} -1 ${ISALIST_${CST}}
COMMAND $<TARGET_FILE:${TARGET_MKUNROLL}> unroll0.cpp.in ${LT} ${CST} - ${ISALIST_${CST}}
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
DEPENDS ${TARGET_MKUNROLL} unroll0.cpp.in.copied
)
Expand All @@ -348,6 +375,26 @@ foreach(T ${LIST_SUPPORTED_FPTYPE})
)
add_custom_target(unroll_target_${ST}_${I} DEPENDS ${UNROLL_TARGET_${CST}_${I}})
endforeach()

#

foreach(I ${LISTSHIFTSTR2})
foreach(E ${ISALIST_${CST}}) # E is "sse2dp"
foreach(N ${NLIST})
string(CONCAT UC unroll2_ ${N} _ ${E} _ ${I} ".cpp") # UC is "unroll2_0_sse2dp_1.cpp"
set(UNROLL2_TARGET_${CST}_${I} ${UNROLL2_TARGET_${CST}_${I}} ${UC})
endforeach()
endforeach()
message(STATUS "Unroll2 target for ${CST}_${I} : ${UNROLL2_TARGET_${CST}_${I}}")

add_custom_command(OUTPUT ${UNROLL2_TARGET_${CST}_${I}}
COMMENT "Generating ${UNROLL2_TARGET_${CST}_${I}}"
COMMAND $<TARGET_FILE:${TARGET_MKUNROLL}> unroll2.cpp.in ${LT} ${CST} -${I} ${ISALIST_${CST}}
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
DEPENDS ${TARGET_MKUNROLL} unroll2.cpp.in.copied
)
add_custom_target(unroll2_target_${ST}_${I} DEPENDS ${UNROLL2_TARGET_${CST}_${I}})
endforeach()
endif()
endforeach()

Expand All @@ -360,7 +407,7 @@ foreach(T ${LIST_SUPPORTED_FPTYPE})

foreach(E ${ISALIST_${CST}}) # E is "sse2dp"
foreach(N ${NLIST})
string(CONCAT U unroll_ ${N} _ ${E}) # U is "unroll_0_sse2dp"
string(CONCAT U unroll_ ${N} _ ${E}) # U is "unroll_0_sse2dp"
string(CONCAT UG ${U} "_obj") # UG is "unroll_0_sse2dp_obj"
string(CONCAT UC ${U} ".cpp") # UC is "unroll_0_sse2dp.cpp"
add_library(${UG} OBJECT ${UC})
Expand All @@ -372,15 +419,28 @@ foreach(T ${LIST_SUPPORTED_FPTYPE})
list(APPEND UNROLL_OBJECTS $<TARGET_OBJECTS:${UG}>)

foreach(I ${LISTSHIFTSTR})
string(CONCAT U unroll_ ${N} _ ${E} _ ${I}) # U is "unroll_0_sse2dp_1"
string(CONCAT U unroll_ ${N} _ ${E} _ ${I}) # U is "unroll_0_sse2dp_1"
string(CONCAT UG ${U} "_obj") # UG is "unroll_0_sse2dp_1_obj"
string(CONCAT UC ${U} ".cpp") # UC is "unroll_0_sse2dp_1.cpp"
add_library(${UG} OBJECT ${UC})
set_target_properties(${UG} PROPERTIES ${COMMON_TARGET_PROPERTIES})
target_include_directories(${UG} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
target_compile_definitions(${UG} PRIVATE ${COMMON_TARGET_DEFINITIONS} ${MACRODEF_${E}})
target_compile_options(${UG} PRIVATE ${CFLAGS_${E}})
add_dependencies(${UG} ${TARGET_HEADERS} unroll_target_${ST})
add_dependencies(${UG} ${TARGET_HEADERS} unroll_target_${ST}_${I})
list(APPEND UNROLL_OBJECTS $<TARGET_OBJECTS:${UG}>)
endforeach()

foreach(I ${LISTSHIFTSTR2})
string(CONCAT U unroll2_ ${N} _ ${E} _ ${I}) # U is "unroll2_0_sse2dp_1"
string(CONCAT UG ${U} "_obj") # UG is "unroll2_0_sse2dp_1_obj"
string(CONCAT UC ${U} ".cpp") # UC is "unroll2_0_sse2dp_1.cpp"
add_library(${UG} OBJECT ${UC})
set_target_properties(${UG} PROPERTIES ${COMMON_TARGET_PROPERTIES})
target_include_directories(${UG} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
target_compile_definitions(${UG} PRIVATE ${COMMON_TARGET_DEFINITIONS} ${MACRODEF_${E}})
target_compile_options(${UG} PRIVATE ${CFLAGS_${E}})
add_dependencies(${UG} ${TARGET_HEADERS} unroll2_target_${ST}_${I})
list(APPEND UNROLL_OBJECTS $<TARGET_OBJECTS:${UG}>)
endforeach()
endforeach()
Expand Down
Loading