On 15 January 2018 at 16:18, Richard Henderson <r...@twiddle.net> wrote: > On 01/15/2018 06:27 AM, Peter Maydell wrote: >> The code looks OK to me. I'm guessing the sparc guest failure is >> just because we now generate more code for some of the comparison >> cases and that doesn't fit in the buffer... > > I don't recall having seen a sparc failure...
See the backtrace in my reply to your tcg pullrequest email :-) >> We could avoid the annoying "load LE/GE immediates to tempreg" >> extra code by having tcg_out_cmp2() return a flag to tell >> the caller which way round to put the conditions for its two >> conditional ARITH_MOV insns (for setcond2) or which condition >> to use for the branch (brcond2), right? > > Um... we do return a condition. I must have missed a trick or > made a mistake somewhere. Oh, yes, rather than returning "condition and flag to tell caller to invert it" we could just flip which condition we return, couldn't we. thanks -- PMM
