does `make check` pass ? if not, that's a good reason to not merge this change :).
if it does, i've got other fixes pending locally related to shifts, so lemme see if i can find time this weekend to see if those also fix this. -mike On Wed, Mar 14, 2012 at 13:44, Henderson, Stuart <stuart.hender...@analog.com> wrote: > A user recently pointed out the following bug in our simulator: > [#7002] simulator bug with saturating shift > where: > R0 = -4(X); > R1 = R0 <<31 (S); > results in 0 instead of 0x80000000. > > The problem is we're not detecting the sign bit being shifted out of the > range of the register. this happens when we're detecting the real shift > count. we check if it's greater than 16, and if it is, shift the value by 16 > and then decrement the shift count by 16. I've no idea why we do this, > however, as we then go straight on to shift the value the rest of the way. > The only side effect of this being that the shift count is now 16 less than > it should be for when we're doing the overflow check later on. > Unless someone can give me a good reason as to why this makes sense, I plan > to remove it. This will fix the bug and doesn't appear to have any other > side effects, although I'm still testing. > > Thanks, > Stu > > > Index: binutils-2.17/sim/bfin/bfin-sim.c > =================================================================== > --- binutils-2.17/sim/bfin/bfin-sim.c (revision 5800) > +++ binutils-2.17/sim/bfin/bfin-sim.c (working copy) > @@ -760,9 +760,6 @@ > sgn <<= mask_cnt; > masked = val & mask; > > - if (real_cnt > 16) > - new_val <<= 16, real_cnt -= 16; > - > new_val <<= real_cnt; > > masked = new_val & mask; > > > > _______________________________________________ > Toolchain-devel mailing list > Toolchain-devel@blackfin.uclinux.org > https://blackfin.uclinux.org/mailman/listinfo/toolchain-devel _______________________________________________ Toolchain-devel mailing list Toolchain-devel@blackfin.uclinux.org https://blackfin.uclinux.org/mailman/listinfo/toolchain-devel
