Re: [Qemu-devel] [HelenOS-devel] [sparc64] Miscomputed minimum of a group of numbers in sparc64 emulation

[Artyom Tarasenko]

On Fri, Jul 1, 2011 at 5:53 PM, Artyom Tarasenko <address@hidden> wrote:
> On Fri, Jul 1, 2011 at 5:20 PM, Laurent Desnogues
> <address@hidden> wrote:
>> On Fri, Jul 1, 2011 at 5:03 PM, Artyom Tarasenko <address@hidden> wrote:
>> [...]
>>>> I find it odd that udivx is using cpu_cc_src and cpu_cc_src2.  Using
>>>> dedicated local temps seems to fix the issue.
>>>
>>> Do we need to copy cpu_src* to further temps at all? IMHO
>>>
>>> -                        tcg_gen_mov_tl(cpu_cc_src, cpu_src1);
>>> -                        tcg_gen_mov_tl(cpu_cc_src2, cpu_src2);
>>> -                        gen_trap_ifdivzero_tl(cpu_cc_src2);
>>> -                        tcg_gen_divu_i64(cpu_dst, cpu_cc_src, cpu_cc_src2);
>>> +                        gen_trap_ifdivzero_tl(cpu_src2);
>>> +                        tcg_gen_divu_i64(cpu_dst, cpu_src1, cpu_src2);
>>>
>>> should do it. Or cpu_src is what you mean by dedicated?
>>
>> You have to use two local temps here to store cpu_src1
>> and cpu_src2 because gen_trap_ifdivzero_tl uses
>> tcg_gen_brcondi_tl (cf tcg/README comment about
>> local usage and jumps).

Thought on a second thought, it's only critical if the jump is taken,
right? And here if the jump is taken, we don't need  cpu_src1 &
cpu_src2.

> The same story as the recent crash with wrpr %pstate, I see. I should
> have done my homework.
> Thanks for the hint!




-- 
Regards,
Artyom Tarasenko

solaris/sparc under qemu blog: http://tyom.blogspot.com/