[trunk] patch to clean up simulation of insns in df.

[Kenneth Zadeck]

Richard Sandiford wrote:
> Kenneth Zadeck <zadeck@naturalbridge.com> writes:
>   
>> +   It would be trivial to define a similar set of functions that work
>> +   in the forwards direction.  The only changes would be to process
>> +   the uses before the defs and properly rename the functions.  This
>> +   has so far not been necessary.
>>     
>
> But this isn't true, is it?  The comment seems to describe what
> the removed function does, but I think the removed function is wrong.
>
> With forward scanning, a value is born at the point of definition and
> dies after the last use (if any).  The point of death is indicated
> by REG_DEAD and REG_UNUSED notes.
>
> So the forward version of df_simulate_on_insn would do something like:
>
>   - kill all values in REG_DEAD notes
>   - make all defined values live
>   - kill all values in REG_UNUSED notes
>
> Despite what the comment says, I don't think it would look at the
> DF_REF_USES at all.
>
> I'm not saying you should say all that in the comment ;)  But I think
> it's one of those comments that's more misleading than helpful.
>
> Richard
>   
you are right, i will fix it.

kenny