Boundary calculation

Boundary calculation

[Ricardo Biehl Pasquali]

Hello.

I can't get the tricky boundary calculation. Perhaps I'm
missing some obvious thing, or maybe it's really some
circumvent for a specific case that was not commented.

The current calculation for boundary is:

	boundary = buffer_size;
	while (boundary * 2 <= LONG_MAX - buffer_size)
		boundary *= 2;

I've checked out some repositories, specially alsa-driver
in ALSA git, and I realized this calculation exists at
least since 2001. I guess it was Jaroslav who wrote.

Visualization of the calculation (B = buffer_size):

                           LONG_MAX
-------------------------------
\______________/           \__/
     B * 4                  B
\______/
 B * 2
\__/
 B

Why isn't a simpler calculation used instead? e.g.:

	/* closest multiple of buffer size less or equal LONG_MAX */
	boundary = LONG_MAX - LONG_MAX % buffer_size;

                           LONG_MAX
-------------------------------
\__________________________/
           B * 6

Cheers!

	pasquali

Re: Boundary calculation

[Clemens Ladisch]

Ricardo Biehl Pasquali wrote:
> The current calculation for boundary is:
>
> 	boundary = buffer_size;
> 	while (boundary * 2 <= LONG_MAX - buffer_size)
> 		boundary *= 2;
>
> Why isn't a simpler calculation used instead? e.g.:
>
> 	/* closest multiple of buffer size less or equal LONG_MAX */
> 	boundary = LONG_MAX - LONG_MAX % buffer_size;

To ensure that the factor is a power of two.

Originally, that was not really necessary, but the algorithm is
now part of the kernel/userspace interface.


Regards,
Clemens

Re: Boundary calculation

[Takashi Iwai]

On Tue, 17 Jul 2018 08:43:27 +0200,
Clemens Ladisch wrote:
> 
> Ricardo Biehl Pasquali wrote:
> > The current calculation for boundary is:
> >
> > 	boundary = buffer_size;
> > 	while (boundary * 2 <= LONG_MAX - buffer_size)
> > 		boundary *= 2;
> >
> > Why isn't a simpler calculation used instead? e.g.:
> >
> > 	/* closest multiple of buffer size less or equal LONG_MAX */
> > 	boundary = LONG_MAX - LONG_MAX % buffer_size;
> 
> To ensure that the factor is a power of two.
> 
> Originally, that was not really necessary, but the algorithm is
> now part of the kernel/userspace interface.

Right.

I don't think we have strictly ABI issue even if relaxing this, but I
also don't think it would give so much benefit, either.


thanks,

Takashi

Boundary calculation

[Ricardo Biehl Pasquali]

Have you addressed your reply to pasqualirb at gmail.com?
I have not received in my mail neither yours nor Takashi's
reply. I had to copy manually the headers and the message.

Thanks!

Clemens Ladisch wrote:
> To ensure that the factor is a power of two.

Ok, but why LONG_MAX - buffer_size?

Re: Boundary calculation

[Clemens Ladisch]

Ricardo Biehl Pasquali wrote:
> Clemens Ladisch wrote:
>> To ensure that the factor is a power of two.
>
> Ok, but why LONG_MAX - buffer_size?

To ensure that adding the position in the buffer to the current pointer
will not overflow.


Regards,
Clemens