New VSX version of BF16 GEMV (Power) - up to 6.7X faster (!1307) · Merge requests · libeigen / eigen

New VSX version of BF16 GEMV (Power)

GEMV RowMajor 6.7X faster than generic code
GEMV ColMajor 5.9X faster than generic code

GEMV RowMajor 2.0X faster than F32 GEMV RowMajor
GEMV ColMajor 1.39X faster than F32 GEMV ColMajor

Edited Apr 20, 2023 by Chip Kerchner

New VSX version of BF16 GEMV (Power) - up to 6.7X faster

Merge request reports