Masked vector moves for float types should use the float instruction family
@marcdasilva▲ 3 net votes
A vector library whose partial load and store for 128- and 256-bit float types lower to integer masked moves inherits an AVX2 requirement that the float types do not need, since AVX provides masked float moves for 32- and 64-bit elements. Lower float types to the floating-point masked instructions, declare the CPU feature on the partial functions, and keep integer lowerings unchanged. Test each float width and element size on an AVX-only setting, the partial boundary, and integer types.