mdaly@gonzo.eta.com (Mike Daly) (06/13/88)
While digging through the IEEE FP test suite (from UC Berkley), we think we found a couple of errors in the multiply and divide operations. These lines appear unchanged in a version of the suite which was mailed to me (thanks Dave H.). The following lines are of interest: Multiply: 2* >= Ep1d1 1m1 xu E 2* <= Ep1d1 -1m1 xu -E For both of these lines, we expect the exceptions to be "xw", not "xu" Similarly for divide: 2/ >= Ep1d1 2 xu E 2/ > E 1i1 xu E 2/ < -E 1i1 xu -E 2/ > eI1 1i2 xu E Again, the exceptions should be "xw", not "xu". The common feature for all of these test is that the result is a number which is "tiny" (underflowing), and then rounds back to a normalized number. Anyone else notice this problem? Mike Daly ...!eta!swde!mdaly ETA Systems Inc. mdaly@gonzo.eta.com 1450 Energy Park Drive mdaly@woods.eta.com St. Paul, MN 55108 (612) 642-3403