new do-count rtl

new do-count rtl

[Doug Evans]

Hi.

What do folks think of adding a new rtl for iterating a specified number 
of times.

Something like

(do-count nr-times iter-var expr1 [expr2 ...])

For now the result would always be void, and iter-var would always have 
mode INT.
The set of expressions is performed nr-times times and iter-var has 
values 0 ... nr-times - 1.

?

Re: new do-count rtl

[Doug Evans]

Doug Evans wrote:
> Hi.
>
> What do folks think of adding a new rtl for iterating a specified 
> number of times.
>
> Something like
>
> (do-count nr-times iter-var expr1 [expr2 ...])
>
> For now the result would always be void, and iter-var would always 
> have mode INT.
> The set of expressions is performed nr-times times and iter-var has 
> values 0 ... nr-times - 1.
>
> ?
>

Checked in.

2009-08-27  Doug Evans  <dje@sebabeach.org>

        Add do-count rtl function.
        * rtl-c.scm (s-do-count): New function.
        (do-count): New rtl handler.
        * rtl-traverse.scm (-rtx-traverse-iteration): New function.
        (-rtx-make-traverser-table): Add ITERATION.
        * rtl.scm (rtx-env-make-iteration-locals): New function.
        * rtx-funcs.scm (do-count): New rtl function.
        * cpu/play.cpu: Add do-count-test insn.
        * doc/rtl.texi: Add docs on do-count.

Re: new do-count rtl

[Dave Korn]

Doug Evans wrote:
> Doug Evans wrote:
>> Hi.
>>
>> What do folks think of adding a new rtl for iterating a specified
>> number of times.
>>
>> Something like
>>
>> (do-count nr-times iter-var expr1 [expr2 ...])
>>
>> For now the result would always be void, and iter-var would always
>> have mode INT.
>> The set of expressions is performed nr-times times and iter-var has
>> values 0 ... nr-times - 1.
>>
>> ?
>>
> 
> Checked in.

  I don't speak scheme or guile, but I'm learning, so forgive me in advance if
this is a dumb question:  Does this differ from using a .map on an .iota when
generating a parameter to an rtl s-expr?

    cheers,
      DaveK

Re: new do-count rtl

[Doug Evans]

Dave Korn wrote:
> Doug Evans wrote:
>   
>> Doug Evans wrote:
>>     
>>> Hi.
>>>
>>> What do folks think of adding a new rtl for iterating a specified
>>> number of times.
>>>
>>> Something like
>>>
>>> (do-count nr-times iter-var expr1 [expr2 ...])
>>>
>>> For now the result would always be void, and iter-var would always
>>> have mode INT.
>>> The set of expressions is performed nr-times times and iter-var has
>>> values 0 ... nr-times - 1.
>>>
>>> ?
>>>
>>>       
>> Checked in.
>>     
>
>   I don't speak scheme or guile, but I'm learning, so forgive me in advance if
> this is a dumb question:  Does this differ from using a .map on an .iota when
> generating a parameter to an rtl s-expr?
>
>     cheers,
>       DaveK
>
>   

If one knows the number of iterations in advance, yes, they're 
essentially equivalent.  do-count means less rtl, which can be useful, 
but that's it.
[As it turns out I need to be able to make the number of iterations an 
expression, and not just a constant.]

Re: new do-count rtl

[Doug Evans]

Doug Evans wrote:
> Doug Evans wrote:
>> Hi.
>>
>> What do folks think of adding a new rtl for iterating a specified 
>> number of times.
>>
>> Something like
>>
>> (do-count nr-times iter-var expr1 [expr2 ...])
>>
>> For now the result would always be void, and iter-var would always 
>> have mode INT.
>> The set of expressions is performed nr-times times and iter-var has 
>> values 0 ... nr-times - 1.
>>
>> ?
>>
>
> Checked in.
>
> 2009-08-27  Doug Evans  <dje@sebabeach.org>
>
>        Add do-count rtl function.
>        * rtl-c.scm (s-do-count): New function.
>        (do-count): New rtl handler.
>        * rtl-traverse.scm (-rtx-traverse-iteration): New function.
>        (-rtx-make-traverser-table): Add ITERATION.
>        * rtl.scm (rtx-env-make-iteration-locals): New function.
>        * rtx-funcs.scm (do-count): New rtl function.
>        * cpu/play.cpu: Add do-count-test insn.
>        * doc/rtl.texi: Add docs on do-count.
>

fyi,
I updated the patch to support an rtl expression for the number of 
iterations, and swapped the iteration-var and number-of-iterations 
parameters (felt more consistent with sequence/parallel).