operating on individual register bits

operating on individual register bits

[Ben Elliston]

Frequently, I find myself wanting to set and clear individual bits of
hardware registers in rtl.  Perhaps such a mechanism already exists.
If it does, I'm not aware of it.  Here is the canonical example:

	 (set cc.z 1)

versus:

	(set cc (or cc (sll SI 1 10))),

which makes much less sense to the reader.

Is it sensible to extend cgen to give names to individual bits of a
register such that operands on registers can adopt this C/Pascal-style
syntax for accessing individual bits?

Ben

Re: operating on individual register bits

[Johan Rydberg]

Ben Elliston wrote:
> 
> Frequently, I find myself wanting to set and clear individual bits of
> hardware registers in rtl.  Perhaps such a mechanism already exists.
> If it does, I'm not aware of it.  Here is the canonical example:
> 
>          (set cc.z 1)
> 
> versus:
> 
>         (set cc (or cc (sll SI 1 10))),
> 
> which makes much less sense to the reader.

To my knowledge there is no such rtx fns.  Looking at your reply to
the same question asked by me some time ago:

 http://sources.redhat.com/ml/cgen/2001-q1/msg00062.html
 
> Is it sensible to extend cgen to give names to individual bits of a
> register such that operands on registers can adopt this C/Pascal-style
> syntax for accessing individual bits?

That would be great!

regards,
johan

operating on individual register bits

[Doug Evans]

Ben Elliston writes:
 > Frequently, I find myself wanting to set and clear individual bits of
 > hardware registers in rtl.  Perhaps such a mechanism already exists.
 > If it does, I'm not aware of it.  Here is the canonical example:
 > 
 > 	 (set cc.z 1)
 > 
 > versus:
 > 
 > 	(set cc (or cc (sll SI 1 10))),
 > 
 > which makes much less sense to the reader.
 > 
 > Is it sensible to extend cgen to give names to individual bits of a
 > register such that operands on registers can adopt this C/Pascal-style
 > syntax for accessing individual bits?

There's comments regarding this in the code and/or docs.
I wasn't sure how I wanted to go (e.g. follow gcc's rtl support
for this or not) so I punted until there was a need and more thought
could be put in.

It's a reasonable thing to add but it needs to be thought about
(try to find lots of architectures with such support and see what
works best to describe them).