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

Merge pull request #2966 from martin-frbg/issue2964 

Ensure that EXPRECISION is disabled for DYNAMIC_ARCH with TARGET=GENERIC and fix CMAKE DYNAMIC_ARCH builds
This commit is contained in:
Martin Kroeker 2020-11-04 16:02:46 +01:00 committed by GitHub
parent f071245939 8cc73fee98
commit 60ab9c783f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 70 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Makefile.system
View File

@ -93,6 +93,12 @@ endif
ifdef TARGET ifdef TARGET
GETARCH_FLAGS := -DFORCE_$(TARGET) GETARCH_FLAGS := -DFORCE_$(TARGET)
GETARCH_FLAGS += -DUSER_TARGET GETARCH_FLAGS += -DUSER_TARGET
ifeq ($(TARGET), GENERIC)
ifeq ($(DYNAMIC_ARCH), 1)
override NO_EXPRECISION=1
export NO_EXPRECiSION
endif
endif
endif endif
# Force fallbacks for 32bit # Force fallbacks for 32bit

8
cmake/os.cmake
View File

@ -84,6 +84,14 @@ if (X86)
set(NO_EXPRECISION 1) set(NO_EXPRECISION 1)
endif () endif ()
if (DYNAMIC_ARCH)
if (TARGET)
if (${TARGET} STREQUAL "GENERIC")
set(NO_EXPRECISION 1)
endif ()
endif ()
endif ()
if (UTEST_CHECK) if (UTEST_CHECK)
set(CCOMMON_OPT "${CCOMMON_OPT} -DUTEST_CHECK") set(CCOMMON_OPT "${CCOMMON_OPT} -DUTEST_CHECK")
set(SANITY_CHECK 1) set(SANITY_CHECK 1)

30
cmake/prebuild.cmake
View File

@ -139,6 +139,36 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
set(CGEMM3M_UNROLL_N 4) set(CGEMM3M_UNROLL_N 4)
set(ZGEMM3M_UNROLL_M 4) set(ZGEMM3M_UNROLL_M 4)
set(ZGEMM3M_UNROLL_N 4) set(ZGEMM3M_UNROLL_N 4)
elseif ("${TCORE}" STREQUAL "BARCELONA")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_SSE3\n")
elseif ("${TCORE}" STREQUAL "STEAMROLLER")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_SSE3\n")
elseif ("${TCORE}" STREQUAL "EXCAVATOR")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_SSE3\n")
elseif ("${TCORE}" STREQUAL "NEHALEM")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_SSE3\n")
elseif ("${TCORE}" STREQUAL "PRESCOTT")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_SSE3\n")
elseif ("${TCORE}" STREQUAL "SANDYBRIDGE")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_AVX\n")
elseif ("${TCORE}" STREQUAL "HASWELL")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_AVX2\n")
elseif ("${TCORE}" STREQUAL "ZEN")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_AVX2\n")
elseif ("${TCORE}" STREQUAL "SKYLAKEX")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_AVX512\n")
elseif ("${TCORE}" STREQUAL "COOPERLAKE")
file(APPEND ${TARGET_CONF_TEMP}
"#define HAVE_AVX512\n")
elseif ("${TCORE}" STREQUAL "ARMV7") elseif ("${TCORE}" STREQUAL "ARMV7")
file(APPEND ${TARGET_CONF_TEMP} file(APPEND ${TARGET_CONF_TEMP}
"#define L1_DATA_SIZE\t65536\n" "#define L1_DATA_SIZE\t65536\n"

28
cmake/system.cmake
View File

@ -64,12 +64,36 @@ if (DEFINED TARGET)
if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU") if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
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 4.7 OR ${GCC_VERSION} VERSION_EQUAL 4.7) if (${GCC_VERSION} VERSION_GREATER 4.7 OR ${GCC_VERSION} VERSION_EQUAL 4.7)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx2")
endif() endif()
elseif (${CMAKE_C_COMPILER_ID} STREQUAL "CLANG") elseif (${CMAKE_C_COMPILER_ID} STREQUAL "CLANG")
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse -msse3 -mavx2")
endif() endif()
endif() endif()
if (${TARGET} STREQUAL "HASWELL" AND NOT NO_AVX2)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx2")
endif()
if (${TARGET} STREQUAL "ZEN" AND NOT NO_AVX2)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx2")
endif()
if (${TARGET} STREQUAL "SANDYBRIDGE" AND NOT NO_AVX)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3 -mavx")
endif()
if (${TARGET} STREQUAL "BARCELONA" OR ${TARGET} STREQUAL "STEAMROLLER" OR ${TARGET} STREQUAL "BULLDOZER" OR ${TARGET} STREQUAL "EXCAVATOR")
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
endif()
if (${TARGET} STREQUAL "PILEDRIVER" OR ${TARGET} STREQUAL "BOBCAT" OR ${TARGET} STREQUAL "OPTERON_SSE3")
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
endif()
if (${TARGET} STREQUAL "PRESCOTT" OR ${TARGET} STREQUAL "NANO")
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
endif()
if (${TARGET} STREQUAL "NEHALEM" OR ${TARGET} STREQUAL "ATOM")
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
endif()
if (${TARGET} STREQUAL "CORE2" OR ${TARGET} STREQUAL "PENRYN" OR ${TARGET} STREQUAL "DUNNINGTON")
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
endif()
if (DEFINED HAVE_SSE) if (DEFINED HAVE_SSE)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
endif() endif()