by Stuart Longland

[permalink] [raw]

Subject: Re: [PATCH] ASoC: Add new TI TLV320AIC3204 CODEC driver

On Sat, Jun 19, 2010 at 02:12:21AM +0100, Mark Brown wrote:
> On Sat, Jun 19, 2010 at 08:24:36AM +1000, Stuart Longland wrote:
> > + /* Page 1 */
> > + if (page == 1) {
> > + if (reg <= 4)
> > + return 1;
>
> I can't help but think that this'd be more legible with switch ()
> statements (GCC has an extension for ranges in switch statements which
> you could use).

I gave this some further thought... I'm not certain that a switch
statement is going to be much clearer. There are two ways I can tackle
this.

One is to go on a page-by-page basis, which is how I do it using the if
statements. Here; I define my ranges so that I start from the very
end... anything beyond page 70 is invalid ... voila, I eliminate those
early on. A number of pages have a similar register pattern, and so I
make use of nested if statements to explain this. The if block for
pages following always use the block before to define the upper,
non-inclusive bound.

The register tests start from register 0. I could perhaps reverse the
outer ifs to start at page 0 and work forwards too... but I instead work
backwards from page 70.

I exit the function as early as possible to skip unneeded checks, as
soon as I know a range is valid or not, I return 1 or 0. Perhaps the 1
or 0 could be made clearer (a couple of #defines maybe?) but to me, it
looks fairly clear.

I could use switch statements to replace some or all of the if
statements. There'd be a small benefit I suppose in making the outer if
statements a switch, but little anywhere else from what I can see.

The other way is that I ignore pages completely; and use the
AIC3204_PGREG macro to define ranges of absolute register addresses.
This may have a small benefit in speed since these are compiled in... as
opposed to runtime masking/shifting, but I don't see that being much
clearer either. I'd still come to a case statement, then return.

This is a function largely intended for debugging, in fact, I'm thinking
I should probably wrap it in #ifdef CONFIG_DEBUG_FS, since the function
isn't called unless debugfs is enabled. So I'm not certain that
performance is worth chasing here given the intended purpose -- it's not
something that's called all the time, nor something that will be used in
a production environment.

That's my thoughts on the issue, perhaps naïve, but I'm not sure
there's any real gain in refactoring this.
--
Stuart Longland (aka Redhatter, VK4MSL) .'''.
Gentoo Linux/MIPS Cobalt and Docs Developer '.'` :
. . . . . . . . . . . . . . . . . . . . . . .'.'
http://dev.gentoo.org/~redhatter :.'

I haven't lost my mind...
...it's backed up on a tape somewhere.

2010-06-22 10:15:05

by Mark Brown

[permalink] [raw]

Subject: Re: [PATCH] ASoC: Add new TI TLV320AIC3204 CODEC driver

On Tue, Jun 22, 2010 at 09:43:14AM +1000, Stuart Longland wrote:
> On Sat, Jun 19, 2010 at 02:12:21AM +0100, Mark Brown wrote:
> > On Sat, Jun 19, 2010 at 08:24:36AM +1000, Stuart Longland wrote:
> > > + /* Page 1 */
> > > + if (page == 1) {
> > > + if (reg <= 4)
> > > + return 1;

> > I can't help but think that this'd be more legible with switch ()
> > statements (GCC has an extension for ranges in switch statements which
> > you could use).

> One is to go on a page-by-page basis, which is how I do it using the if
> statements. Here; I define my ranges so that I start from the very
> end... anything beyond page 70 is invalid ... voila, I eliminate those
> early on. A number of pages have a similar register pattern, and so I
> make use of nested if statements to explain this. The if block for
> pages following always use the block before to define the upper,
> non-inclusive bound.

It's not so much the outer ifs that were bothering me, it's the inner
ones where you're doing the final register ranges as just a sequence of
if statements (not even if ... else) which really bothered me here. The
code just doesn't look like what it's trying to do.

> This is a function largely intended for debugging, in fact, I'm thinking
> I should probably wrap it in #ifdef CONFIG_DEBUG_FS, since the function
> isn't called unless debugfs is enabled. So I'm not certain that
> performance is worth chasing here given the intended purpose -- it's not
> something that's called all the time, nor something that will be used in
> a production environment.

Oh, I thought you were using it to filter the CODEC register displays?

> That's my thoughts on the issue, perhaps naïve, but I'm not sure
> there's any real gain in refactoring this.

It's fairly hard to read at the minute -