floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add support for compiling SVE kernels with the NVIDIA HPC compiler

This commit is contained in:
Martin Kroeker 2023-08-25 17:11:04 +02:00 committed by GitHub
parent 8794544b43
commit 49689fbef7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 34 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
cmake/cc.cmake
View File

@ -180,6 +180,9 @@ endif ()
if (${CORE} STREQUAL NEOVERSEN2) if (${CORE} STREQUAL NEOVERSEN2)
if (NOT DYNAMIC_ARCH) if (NOT DYNAMIC_ARCH)
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
set (CCOMMON_OPT "${CCOMMON_OPT} -Msve_intrinsics -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
else ()
execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION) execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4) if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2") set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
@ -188,9 +191,13 @@ if (${CORE} STREQUAL NEOVERSEN2)
endif() endif()
endif () endif ()
endif () endif ()
endif ()
if (${CORE} STREQUAL NEOVERSEV1) if (${CORE} STREQUAL NEOVERSEV1)
if (NOT DYNAMIC_ARCH) if (NOT DYNAMIC_ARCH)
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
set (CCOMMON_OPT "${CCOMMON_OPT} -Msve_intrinsics -march=armv8.4-a+sve -mtune=neoverse-v1")
else ()
execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION) execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4) if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.4-a+sve -mtune=neoverse-v1") set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.4-a+sve -mtune=neoverse-v1")
@ -199,6 +206,7 @@ if (${CORE} STREQUAL NEOVERSEV1)
endif() endif()
endif() endif()
endif () endif ()
endif ()
if (${CORE} STREQUAL NEOVERSEN1) if (${CORE} STREQUAL NEOVERSEN1)
if (NOT DYNAMIC_ARCH) if (NOT DYNAMIC_ARCH)
@ -213,9 +221,13 @@ endif ()
if (${CORE} STREQUAL ARMV8SVE) if (${CORE} STREQUAL ARMV8SVE)
if (NOT DYNAMIC_ARCH) if (NOT DYNAMIC_ARCH)
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
set (CCOMMON_OPT "${CCOMMON_OPT} -Msve_intrinsics -march=armv8-a+sve")
else ()
set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8-a+sve") set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8-a+sve")
endif () endif ()
endif () endif ()
endif ()
if (${CORE} STREQUAL CORTEXA510) if (${CORE} STREQUAL CORTEXA510)
if (NOT DYNAMIC_ARCH) if (NOT DYNAMIC_ARCH)

12
cmake/system.cmake
View File

@ -282,6 +282,9 @@ if (DEFINED TARGET)
endif() endif()
if (${TARGET} STREQUAL NEOVERSEV1) if (${TARGET} STREQUAL NEOVERSEV1)
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -Msve_intrinsics -march=armv8.4-a+sve -mtune=neoverse-v1")
else ()
execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION) execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4) if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.4-a+sve -mtune=neoverse-v1") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.4-a+sve -mtune=neoverse-v1")
@ -289,7 +292,11 @@ if (${TARGET} STREQUAL NEOVERSEV1)
message(FATAL_ERROR "Compiler ${CMAKE_C_COMPILER} ${GCC_VERSION} does not support Neoverse V1.") message(FATAL_ERROR "Compiler ${CMAKE_C_COMPILER} ${GCC_VERSION} does not support Neoverse V1.")
endif() endif()
endif() endif()
endif()
if (${TARGET} STREQUAL NEOVERSEN2) if (${TARGET} STREQUAL NEOVERSEN2)
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -Msve-intrinsics -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
else ()
execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION) execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4) if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
@ -297,9 +304,14 @@ if (${TARGET} STREQUAL NEOVERSEV1)
message(FATAL_ERROR "Compiler $${CMAKE_C_COMPILER} {GCC_VERSION} does not support Neoverse N2.") message(FATAL_ERROR "Compiler $${CMAKE_C_COMPILER} {GCC_VERSION} does not support Neoverse N2.")
endif() endif()
endif() endif()
endif()
if (${TARGET} STREQUAL ARMV8SVE) if (${TARGET} STREQUAL ARMV8SVE)
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -Msve-intrinsics -march=armv8.2-a+sve")
else ()
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.2-a+sve") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.2-a+sve")
endif() endif()
endif()
endif() endif()