Re: [Qemu-devel] [PATCH] target-m68k: Implement bfffo

[Laurent Vivier]

Le 15/11/2016 à 22:21, Richard Henderson a écrit :
> On 11/15/2016 10:07 PM, Laurent Vivier wrote:
>> Le 15/11/2016 à 21:44, Richard Henderson a écrit :
>>> Signed-off-by: Richard Henderson <address@hidden>
>>> ---
>>>
>>> I've started a glibc test run with this, but I don't expect overmuch.
>>> The only applications I can see are "bfffo *,0,32,dN" which isn't
>>> exactly exhaustive.  Probably better to hand craft some tests vs real
>>> hardware.
>>>
>>> Considering the prevelance of small helper functions calling ctz/clz,
>>> I'm thinking of adding an opcode to tcg for this.  Certainly all of
>>> the common hosts support it natively...
>>
>> I've booted an etch-m68k container with patch and it works.
> 
> Thanks.
> 
>>> @@ -3943,11 +3960,17 @@ DISAS_INSN(bfop_reg)
>>>              tcg_gen_rotl_i32(QREG_CC_N, src, tmp);
>>>              tcg_gen_andc_i32(QREG_CC_N, QREG_CC_N, mask);
>>>              tcg_gen_rotr_i32(mask, mask, tmp);
>>> +            if (!TCGV_IS_UNUSED(tofs)) {
>>> +                tcg_gen_mov_tl(tofs, tmp);
>>
>> _tl suffix is never used with m68k, should we?
>>
> 
> Gah.  Typing habit from other targets.
> 
> I don't think there's any point in using _tl for m68k, as there will
> never be a 64-bit version.  If you could fix that as you apply to your
> tree, that would be great.

I'm going to add the three bitfield patches to the m68k-for-2.9 branch,
fixing this.

Thanks,
Laurent