On 05/16/2018 03:17 PM, Kalamatee wrote: > Hi, > > After hunting out a problem using the softloat code on m68k, with the > assistance of the WinUAE author (Toni Wilen) we think we have noticed a bug > dating back to 1994. > > Laddsf$nf returns values with the wrong sign, because it clears the sign > bit, before caching the wrong value and then attempting to use it. > > the movel d0,d7 is either in the wrong place (should be before the sign bit > is cleared), or it should be removed completely and the following andl IMM > (0x80000000),d7 should be changed to use d2 instead (which should have the > correct sign from the prior eorl) It would really help if you could provide the input values that trigger this erroneous path. It helps with initial validation of the change as well as long term as we can add it to the regression suite.
Jeff
