From: Jesse Gross <[email protected]>
On most architectures division is an expensive operation and accessing an
element currently requires four of them. This performance penalty
effectively precludes flex arrays from being used on any kind of fast
path. However, two of these divisions can be handled at creation time and
the others can be replaced by a reciprocal divide, completely avoiding
real divisions on access.
Signed-off-by: Jesse Gross <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Eric Paris <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
[rebase on top of changes to support 0 len elements: eparis]
[initialize part_nr when array fits entirely in base: eparis]
Signed-off-by: Eric Paris <[email protected]>
---
include/linux/flex_array.h | 2 +
lib/flex_array.c | 51 +++++++++++++++++++++++++-------------------
2 files changed, 31 insertions(+), 22 deletions(-)
diff --git a/include/linux/flex_array.h b/include/linux/flex_array.h
index ebeb2f3..6843cf1 100644
--- a/include/linux/flex_array.h
+++ b/include/linux/flex_array.h
@@ -21,6 +21,8 @@ struct flex_array {
struct {
int element_size;
int total_nr_elements;
+ int elems_per_part;
+ u32 reciprocal_elems;
struct flex_array_part *parts[];
};
/*
diff --git a/lib/flex_array.c b/lib/flex_array.c
index cab7621..9b8b894 100644
--- a/lib/flex_array.c
+++ b/lib/flex_array.c
@@ -24,6 +24,7 @@
#include <linux/slab.h>
#include <linux/stddef.h>
#include <linux/module.h>
+#include <linux/reciprocal_div.h>
struct flex_array_part {
char elements[FLEX_ARRAY_PART_SIZE];
@@ -70,15 +71,15 @@ static inline int elements_fit_in_base(struct flex_array *fa)
* Element size | Objects | Objects |
* PAGE_SIZE=4k | 32-bit | 64-bit |
* ---------------------------------|
- * 1 bytes | 4186112 | 2093056 |
- * 2 bytes | 2093056 | 1046528 |
- * 3 bytes | 1395030 | 697515 |
- * 4 bytes | 1046528 | 523264 |
- * 32 bytes | 130816 | 65408 |
- * 33 bytes | 126728 | 63364 |
- * 2048 bytes | 2044 | 1022 |
- * 2049 bytes | 1022 | 511 |
- * void * | 1046528 | 261632 |
+ * 1 bytes | 4177920 | 2088960 |
+ * 2 bytes | 2088960 | 1044480 |
+ * 3 bytes | 1392300 | 696150 |
+ * 4 bytes | 1044480 | 522240 |
+ * 32 bytes | 130560 | 65408 |
+ * 33 bytes | 126480 | 63240 |
+ * 2048 bytes | 2040 | 1020 |
+ * 2049 bytes | 1020 | 510 |
+ * void * | 1044480 | 261120 |
*
* Since 64-bit pointers are twice the size, we lose half the
* capacity in the base structure. Also note that no effort is made
@@ -88,11 +89,15 @@ struct flex_array *flex_array_alloc(int element_size, unsigned int total,
gfp_t flags)
{
struct flex_array *ret;
+ int elems_per_part = 0;
+ int reciprocal_elems = 0;
int max_size = 0;
- if (element_size)
- max_size = FLEX_ARRAY_NR_BASE_PTRS *
- FLEX_ARRAY_ELEMENTS_PER_PART(element_size);
+ if (element_size) {
+ elems_per_part = FLEX_ARRAY_ELEMENTS_PER_PART(element_size);
+ reciprocal_elems = reciprocal_value(elems_per_part);
+ max_size = FLEX_ARRAY_NR_BASE_PTRS * elems_per_part;
+ }
/* max_size will end up 0 if element_size > PAGE_SIZE */
if (total > max_size)
@@ -102,6 +107,8 @@ struct flex_array *flex_array_alloc(int element_size, unsigned int total,
return NULL;
ret->element_size = element_size;
ret->total_nr_elements = total;
+ ret->elems_per_part = elems_per_part;
+ ret->reciprocal_elems = reciprocal_elems;
if (elements_fit_in_base(ret) && !(flags & __GFP_ZERO))
memset(&ret->parts[0], FLEX_ARRAY_FREE,
FLEX_ARRAY_BASE_BYTES_LEFT);
@@ -112,7 +119,7 @@ EXPORT_SYMBOL(flex_array_alloc);
static int fa_element_to_part_nr(struct flex_array *fa,
unsigned int element_nr)
{
- return element_nr / FLEX_ARRAY_ELEMENTS_PER_PART(fa->element_size);
+ return reciprocal_divide(element_nr, fa->reciprocal_elems);
}
/**
@@ -141,12 +148,12 @@ void flex_array_free(struct flex_array *fa)
EXPORT_SYMBOL(flex_array_free);
static unsigned int index_inside_part(struct flex_array *fa,
- unsigned int element_nr)
+ unsigned int element_nr,
+ unsigned int part_nr)
{
unsigned int part_offset;
- part_offset = element_nr %
- FLEX_ARRAY_ELEMENTS_PER_PART(fa->element_size);
+ part_offset = element_nr - part_nr * fa->elems_per_part;
return part_offset * fa->element_size;
}
@@ -186,7 +193,7 @@ __fa_get_part(struct flex_array *fa, int part_nr, gfp_t flags)
int flex_array_put(struct flex_array *fa, unsigned int element_nr, void *src,
gfp_t flags)
{
- int part_nr;
+ int part_nr = 0;
struct flex_array_part *part;
void *dst;
@@ -202,7 +209,7 @@ int flex_array_put(struct flex_array *fa, unsigned int element_nr, void *src,
if (!part)
return -ENOMEM;
}
- dst = &part->elements[index_inside_part(fa, element_nr)];
+ dst = &part->elements[index_inside_part(fa, element_nr, part_nr)];
memcpy(dst, src, fa->element_size);
return 0;
}
@@ -217,7 +224,7 @@ EXPORT_SYMBOL(flex_array_put);
*/
int flex_array_clear(struct flex_array *fa, unsigned int element_nr)
{
- int part_nr;
+ int part_nr = 0;
struct flex_array_part *part;
void *dst;
@@ -233,7 +240,7 @@ int flex_array_clear(struct flex_array *fa, unsigned int element_nr)
if (!part)
return -EINVAL;
}
- dst = &part->elements[index_inside_part(fa, element_nr)];
+ dst = &part->elements[index_inside_part(fa, element_nr, part_nr)];
memset(dst, FLEX_ARRAY_FREE, fa->element_size);
return 0;
}
@@ -302,7 +309,7 @@ EXPORT_SYMBOL(flex_array_prealloc);
*/
void *flex_array_get(struct flex_array *fa, unsigned int element_nr)
{
- int part_nr;
+ int part_nr = 0;
struct flex_array_part *part;
if (!fa->element_size)
@@ -317,7 +324,7 @@ void *flex_array_get(struct flex_array *fa, unsigned int element_nr)
if (!part)
return NULL;
}
- return &part->elements[index_inside_part(fa, element_nr)];
+ return &part->elements[index_inside_part(fa, element_nr, part_nr)];
}
EXPORT_SYMBOL(flex_array_get);
--
1.7.1
On Mon, 2011-05-02 at 13:26 -0400, Eric Paris wrote:
> On most architectures division is an expensive operation and accessing an
> element currently requires four of them. This performance penalty
> effectively precludes flex arrays from being used on any kind of fast
> path. However, two of these divisions can be handled at creation time and
> the others can be replaced by a reciprocal divide, completely avoiding
> real divisions on access.
>
> Signed-off-by: Jesse Gross <[email protected]>
> Cc: David Rientjes <[email protected]>
> Cc: Eric Paris <[email protected]>
> Signed-off-by: Andrew Morton <[email protected]>
> [rebase on top of changes to support 0 len elements: eparis]
> [initialize part_nr when array fits entirely in base: eparis]
> Signed-off-by: Eric Paris <[email protected]>
This seems sane to me, especially in the element access case.
Acked-by: Dave Hansen <[email protected]>
-- Dave
On Mon, May 2, 2011 at 10:36 AM, Dave Hansen <[email protected]> wrote:
> On Mon, 2011-05-02 at 13:26 -0400, Eric Paris wrote:
>> On most architectures division is an expensive operation and accessing an
>> element currently requires four of them. ?This performance penalty
>> effectively precludes flex arrays from being used on any kind of fast
>> path. ?However, two of these divisions can be handled at creation time and
>> the others can be replaced by a reciprocal divide, completely avoiding
>> real divisions on access.
>>
>> Signed-off-by: Jesse Gross <[email protected]>
>> Cc: David Rientjes <[email protected]>
>> Cc: Eric Paris <[email protected]>
>> Signed-off-by: Andrew Morton <[email protected]>
>> [rebase on top of changes to support 0 len elements: eparis]
>> [initialize part_nr when array fits entirely in base: eparis]
>> Signed-off-by: Eric Paris <[email protected]>
>
> This seems sane to me, especially in the element access case.
>
> Acked-by: Dave Hansen <[email protected]>
Thanks Eric!