Merge pull request #3425 from binebrank/arm_sve_dgemm

Add dgemm kernel for arm64 SVE
This commit is contained in:
Martin Kroeker
2021-11-26 16:14:55 +01:00
committed by GitHub
20 changed files with 4742 additions and 33 deletions

View File

@@ -1198,6 +1198,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#else
#endif
#ifdef FORCE_ARMV8SVE
#define FORCE
#define ARCHITECTURE "ARM64"
#define SUBARCHITECTURE "ARMV8SVE"
#define SUBDIRNAME "arm64"
#define ARCHCONFIG "-DARMV8SVE " \
"-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 " \
"-DL2_SIZE=262144 -DL2_LINESIZE=64 " \
"-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=32 " \
"-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DHAVE_SVE -DARMV8"
#define LIBNAME "armv8sve"
#define CORENAME "ARMV8SVE"
#endif
#ifdef FORCE_ARMV8
#define FORCE
@@ -1436,7 +1450,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
"-DL2_SIZE=8388608 -DL2_LINESIZE=256 -DL2_ASSOCIATIVE=8 " \
"-DL3_SIZE=0 -DL3_LINESIZE=0 -DL3_ASSOCIATIVE=0 " \
"-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
"-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8"
"-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DHAVE_SVE -DARMV8"
#define LIBNAME "a64fx"
#define CORENAME "A64FX"
#else