"""(Triforce, 11-22, the performance audit's LIFT-03, after Source A's) SSE fast paths for the float forms the lifter still writes lane by lane, behind A64_VECFP (off by default). The kit turns straight FADD / FSUB, FMUL .4S and FMUL / FMLA / FMLS .4S by element into one SSE op with the scalar fast path's tests on every lane (a64_lift.py's vectorize_fpx and vectorize_elem). These stay loops of one helper call a lane, or exact integer code a lane: elem FMUL, FMLA, FMLS .2S by element vec FMUL .2S; FMLA and FMLS .2S and .4S; FDIV .2S and .4S minmax FMAX, FMIN, FMAXNM, FMINNM: scalar S, .2S and .4S fcvt FCVT S to D and D to S (scalar); FCVTL and FCVTL2 (.2S to .2D); FCVTN (.2D to .2S) With A64_VECFP=1 (or a list of the groups above) each such instruction is lifted as if (!a64_vfpx_q_