Re: How best to write a series of related test_bit() calls?
From: Geert Uytterhoeven <geert@linux-m68k.org>
Date: 2015-02-25 08:32:41
On Tue, Feb 24, 2015 at 3:58 PM, David Howells [off-list ref] wrote:
I have a situation where I want to do something that boils down to:
if (test_bit(BIT_A, &p->flags) {
...
} else if (test_bit(BIT_B, &p->flags) {
...
} else if (test_bit(BIT_C, &p->flags) {
...
} else if (test_bit(BIT_D, &p->flags) {
...
} else if (test_bit(BIT_E, &p->flags) {
}
Note that all bits are in the same unsigned long and I don't necessarily
expect the contents of p->flags to change whilst I'm looking at it.
Since the address parameter for test_bit() contains a 'volatile' keyword, the
compiler will emit a separate load (or bit test) for each condition when in
fact the most efficient way is almost certainly one load and a bunch of
bitwise-and or bit-test-in-register instructions.
Is it worth adding a __test_bit() that doesn't have the volatile? Or should I
just rewrite it manually as?
unsigned long flags = p->flags;
if (test_bit(BIT_A, &flags) {
...
} else if (test_bit(BIT_B, &flags) {
...
Or even:
unsigned long flags = p->flags;
if ((1 << BIT_A) & flags) {
...
} else if ((1 << BIT_B) & flags) {
...
How about using READ_ONCE() and BIT()?
unsigned long flags = READ_ONCE(p->flags_;
if (BIT(BIT_A) & flags) {
...
} else ...
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds