When dealing with array indices, the parent's index can be obtained
using the formula (i - 1) / 2. However, when working with byte offsets,
this approach is not straightforward. To address this, we have
introduced a branch-free parent function that does not require any
division operations to calculate the parent's byte offset.
Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
---
This patch has undergone unit testing using the following code [1].
[1]:
static int test(void)
{
size_t i, p, size, lsbit;
for (i = 0; i < 10000; i++) {
size = get_random_u32() % (1U << 10);
lsbit = size & -size;
i = get_random_u32() % (1U << 20) * size + size;
p = parent(i, lsbit, size);
if (p != (i / size - 1) / 2 * size)
return -1;
}
return 0;
}
fs/bcachefs/util.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/fs/bcachefs/util.c b/fs/bcachefs/util.c
index bbc83b43162e..f5bbf96df2ce 100644
--- a/fs/bcachefs/util.c
+++ b/fs/bcachefs/util.c
@@ -907,6 +907,13 @@ static inline void do_swap(void *base, size_t n, size_t size,
size);
}
+static inline size_t parent(size_t i, size_t lsbit, size_t size)
+{
+ i -= size;
+ i -= size & -(i & lsbit);
+ return i >> 1;
+}
+
void eytzinger0_sort(void *base, size_t n, size_t size,
int (*cmp_func)(const void *, const void *, size_t),
void (*swap_func)(void *, void *, size_t))
--
2.25.1On Sun, Jan 21, 2024 at 11:36:46PM +0800, Kuan-Wei Chiu wrote:
> When dealing with array indices, the parent's index can be obtained
> using the formula (i - 1) / 2. However, when working with byte offsets,
> this approach is not straightforward. To address this, we have
> introduced a branch-free parent function that does not require any
> division operations to calculate the parent's byte offset.
This is a good commit message - but it would be even better if it was a
function comment on parent()
>
> Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
> ---
> This patch has undergone unit testing using the following code [1].
>
> [1]:
> static int test(void)
> {
> size_t i, p, size, lsbit;
>
> for (i = 0; i < 10000; i++) {
> size = get_random_u32() % (1U << 10);
> lsbit = size & -size;
> i = get_random_u32() % (1U << 20) * size + size;
> p = parent(i, lsbit, size);
> if (p != (i / size - 1) / 2 * size)
> return -1;
> }
>
> return 0;
> }
>
> fs/bcachefs/util.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/fs/bcachefs/util.c b/fs/bcachefs/util.c
> index bbc83b43162e..f5bbf96df2ce 100644
> --- a/fs/bcachefs/util.c
> +++ b/fs/bcachefs/util.c
> @@ -907,6 +907,13 @@ static inline void do_swap(void *base, size_t n, size_t size,
> size);
> }
>
> +static inline size_t parent(size_t i, size_t lsbit, size_t size)
> +{
> + i -= size;
> + i -= size & -(i & lsbit);
> + return i >> 1;
> +}
> +
> void eytzinger0_sort(void *base, size_t n, size_t size,
> int (*cmp_func)(const void *, const void *, size_t),
> void (*swap_func)(void *, void *, size_t))
> --
> 2.25.1
>
On Sun, Jan 21, 2024 at 11:17:30AM -0500, Kent Overstreet wrote:
> On Sun, Jan 21, 2024 at 11:36:46PM +0800, Kuan-Wei Chiu wrote:
> > When dealing with array indices, the parent's index can be obtained
> > using the formula (i - 1) / 2. However, when working with byte offsets,
> > this approach is not straightforward. To address this, we have
> > introduced a branch-free parent function that does not require any
> > division operations to calculate the parent's byte offset.
>
> This is a good commit message - but it would be even better if it was a
> function comment on parent()
>
Sure, however, it seems that sort_cmp_size() can be directly replaced
with the sort function from include/linux. Once we decide on the
cleanup tasks, if we still choose to retain this patch, I will make the
adjustments.
> >
> > Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
> > ---
> > This patch has undergone unit testing using the following code [1].
> >
> > [1]:
> > static int test(void)
> > {
> > size_t i, p, size, lsbit;
> >
> > for (i = 0; i < 10000; i++) {
> > size = get_random_u32() % (1U << 10);
> > lsbit = size & -size;
> > i = get_random_u32() % (1U << 20) * size + size;
> > p = parent(i, lsbit, size);
> > if (p != (i / size - 1) / 2 * size)
> > return -1;
> > }
> >
> > return 0;
> > }
> >
> > fs/bcachefs/util.c | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/fs/bcachefs/util.c b/fs/bcachefs/util.c
> > index bbc83b43162e..f5bbf96df2ce 100644
> > --- a/fs/bcachefs/util.c
> > +++ b/fs/bcachefs/util.c
> > @@ -907,6 +907,13 @@ static inline void do_swap(void *base, size_t n, size_t size,
> > size);
> > }
> >
> > +static inline size_t parent(size_t i, size_t lsbit, size_t size)
> > +{
> > + i -= size;
> > + i -= size & -(i & lsbit);
> > + return i >> 1;
> > +}
> > +
> > void eytzinger0_sort(void *base, size_t n, size_t size,
> > int (*cmp_func)(const void *, const void *, size_t),
> > void (*swap_func)(void *, void *, size_t))
> > --
> > 2.25.1
> >
On Mon, Jan 22, 2024 at 01:05:28AM +0800, Kuan-Wei Chiu wrote: > On Sun, Jan 21, 2024 at 11:17:30AM -0500, Kent Overstreet wrote: > > On Sun, Jan 21, 2024 at 11:36:46PM +0800, Kuan-Wei Chiu wrote: > > > When dealing with array indices, the parent's index can be obtained > > > using the formula (i - 1) / 2. However, when working with byte offsets, > > > this approach is not straightforward. To address this, we have > > > introduced a branch-free parent function that does not require any > > > division operations to calculate the parent's byte offset. > > > > This is a good commit message - but it would be even better if it was a > > function comment on parent() > > > Sure, however, it seems that sort_cmp_size() can be directly replaced > with the sort function from include/linux. Once we decide on the > cleanup tasks, if we still choose to retain this patch, I will make the > adjustments. nice catch - looks like sort_r() is the more recent addition, so that's how that happened.
© 2016 - 2025 Red Hat, Inc.