Narrowing converting instruction mismatch issue for number infinity

[Shafi10x]

If a number is infinity in FP64, after converting it into FP32, will it be infinity or of max magnitude? According to spike, it is infinity but according to RTL implementation, it is of max magnitude.Actually, overflow and infinity are handled together i.e saturating it or making it max magnitude. I think it is fine to handle overflow like this, but for infinity, there should be another case i.e infinity should remain infinity.

[pascalgouedo]

In principle I would agree that infinity stays infinity whatever operation happens to it...

[stmach]

Indeed, inf should always map to inf, no matter the rounding mode or format in use. Under which circumstances do you observe max magnitude being returned? A quick glance into the RTL indicates that this issue should only happen when casting infs with round mode RDN or RTZ (which should be fixed, of course).

[Shafi10x]

In my case it was RTZ mode. I think the following part could be fixed.

      // Overflow or infinities (for proper rounding)
      if ((destination_exp_q >= signed'(2**fpnew_pkg::exp_bits(dst_fmt_q2))-1) ||
          (~src_is_int_q && info_q.is_inf)) begin
        final_exp       = unsigned'(2**fpnew_pkg::exp_bits(dst_fmt_q2)-2); // largest normal value
        preshift_mant   = '1;                           // largest normal value and RS bits set
        of_before_round = 1'b1;