call fp16<->fp32 routines in fbgemm from Half2Float and Float2Half operators (#30715)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30715
Changed caffe2/caffe2/TARGETS file to define USE_FBGEMM for x86 and USE_SSE_ONLY is not defined.
Test Plan: buck test caffe2/caffe2:caffe2_test_cpu -- Float16
Reviewed By: jianyuh
Differential Revision: D18806067
fbshipit-source-id: 1b44b90a9f6dc3c27f81a46038c0f7542ed2bab3
