merge reducemax(float16) reducesum(float16) div(float16) negate(float16) multiplydidm(float16) matrixmul(all,float16/int8) sum(float16/int/int8) sign/sub/sumdim/subdim( float16)