There is no option to perform 64bit integer sqrt on 32bit platform.
Added stronger typed int_sqrt64 enables the 64bit calculations to
be performed on 32bit platforms. Although int_sqrt() is a rough
approximation, the same algorithm is used in int_sqrt64() as good
enough on 32bit platform.
Signed-off-by: Crt Mori <[email protected]>
---
include/linux/kernel.h | 1 +
lib/int_sqrt.c | 26 ++++++++++++++++++++++++++
2 files changed, 27 insertions(+)
diff --git a/include/linux/kernel.h b/include/linux/kernel.h
index 0ad4c3044cf9..09f53063f2c9 100644
--- a/include/linux/kernel.h
+++ b/include/linux/kernel.h
@@ -459,6 +459,7 @@ extern int kernel_text_address(unsigned long addr);
extern int func_ptr_is_kernel_text(void *ptr);
unsigned long int_sqrt(unsigned long);
+u64 int_sqrt64(u64 x);
extern void bust_spinlocks(int yes);
extern int oops_in_progress; /* If set, an oops, panic(), BUG() or die() is in progress */
diff --git a/lib/int_sqrt.c b/lib/int_sqrt.c
index 1ef4cc344977..de5836448ea3 100644
--- a/lib/int_sqrt.c
+++ b/lib/int_sqrt.c
@@ -36,3 +36,29 @@ unsigned long int_sqrt(unsigned long x)
return y;
}
EXPORT_SYMBOL(int_sqrt);
+
+/**
+ * int_sqrt64 - strongly typed int_sqrt function
+ * @x: 64bit integer of which to calculate the sqrt
+ */
+u64 int_sqrt64(u64 x)
+{
+ u64 b, m, y = 0;
+
+ if (x <= 1)
+ return x;
+
+ m = 1ULL << (64 - 2);
+ while (m != 0) {
+ b = y + m;
+ y >>= 1;
+
+ if (x >= b) {
+ x -= b;
+ y += m;
+ }
+ m >>= 2;
+ }
+ return y;
+}
+EXPORT_SYMBOL(int_sqrt64);
--
2.15.0
On Mon, 2017-12-18 at 16:05 +0100, Crt Mori wrote:
> There is no option to perform 64bit integer sqrt on 32bit platform.
> Added stronger typed int_sqrt64 enables the 64bit calculations to
> be performed on 32bit platforms. Although int_sqrt() is a rough
> approximation, the same algorithm is used in int_sqrt64() as good
> enough on 32bit platform.
[]
> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
[]
> @@ -459,6 +459,7 @@ extern int kernel_text_address(unsigned long addr);
> extern int func_ptr_is_kernel_text(void *ptr);
>
> unsigned long int_sqrt(unsigned long);
> +u64 int_sqrt64(u64 x);
>
> extern void bust_spinlocks(int yes);
> extern int oops_in_progress; /* If set, an oops, panic(), BUG() or die() is in progress */
> diff --git a/lib/int_sqrt.c b/lib/int_sqrt.c
[]
> @@ -36,3 +36,29 @@ unsigned long int_sqrt(unsigned long x)
> return y;
> }
> EXPORT_SYMBOL(int_sqrt);
> +
> +/**
> + * int_sqrt64 - strongly typed int_sqrt function
> + * @x: 64bit integer of which to calculate the sqrt
> + */
> +u64 int_sqrt64(u64 x)
> +{
> + u64 b, m, y = 0;
> +
> + if (x <= 1)
> + return x;
> +
> + m = 1ULL << (64 - 2);
> + while (m != 0) {
> + b = y + m;
> + y >>= 1;
> +
> + if (x >= b) {
> + x -= b;
> + y += m;
> + }
> + m >>= 2;
> + }
> + return y;
> +}
> +EXPORT_SYMBOL(int_sqrt64);
int_sqrt is now optimized with __fls
Because u64 is unsigned long on 64 bit arches, this
could probably be optimized for 64 bit distributions
with something like:
#if BITS_PER_LONG != 64
#define int_sqrt64 _int_sqrt64
u64 _int_sqrt64(u64 x);
#else
#define int_sqrt64(x) int_sqrt
#endif
...
#if BITS_PER_LONG != 64
u64 _int_sqrt64(u64 x)
{ etc.. }
#endif
On Mon, Dec 18, 2017 at 04:05:44PM +0100, Crt Mori wrote:
> There is no option to perform 64bit integer sqrt on 32bit platform.
> Added stronger typed int_sqrt64 enables the 64bit calculations to
> be performed on 32bit platforms. Although int_sqrt() is a rough
> approximation, the same algorithm is used in int_sqrt64() as good
> enough on 32bit platform.
You clearly haven't read a recent version of the file you're patching.
Please take a moment to do so now.
> +/**
> + * int_sqrt64 - strongly typed int_sqrt function
> + * @x: 64bit integer of which to calculate the sqrt
> + */
> +u64 int_sqrt64(u64 x)
Please explain how the result of sqrt(u64) can be larger than u32.
Also, I expect that this fact could be exploited to optimize this for
32bit archs if one were so inclined.
> +{
> + u64 b, m, y = 0;
> +
> + if (x <= 1)
> + return x;
> +
> + m = 1ULL << (64 - 2);
> + while (m != 0) {
> + b = y + m;
> + y >>= 1;
> +
> + if (x >= b) {
> + x -= b;
> + y += m;
> + }
> + m >>= 2;
> + }
> + return y;
> +}
so yeah, no, please try again after reading the current file.
On 18 December 2017 at 17:44, Peter Zijlstra <[email protected]> wrote:
> On Mon, Dec 18, 2017 at 04:05:44PM +0100, Crt Mori wrote:
>> There is no option to perform 64bit integer sqrt on 32bit platform.
>> Added stronger typed int_sqrt64 enables the 64bit calculations to
>> be performed on 32bit platforms. Although int_sqrt() is a rough
>> approximation, the same algorithm is used in int_sqrt64() as good
>> enough on 32bit platform.
>
> You clearly haven't read a recent version of the file you're patching.
> Please take a moment to do so now.
>
IIO kernel does not have the recent version in, so thanks for heads
up. It does not change much for my function.
>> +/**
>> + * int_sqrt64 - strongly typed int_sqrt function
>> + * @x: 64bit integer of which to calculate the sqrt
>> + */
>> +u64 int_sqrt64(u64 x)
>
> Please explain how the result of sqrt(u64) can be larger than u32.
>
My hand calculator tells me it could be.
ffffffff×ffffffff = FFFFFFFE00000001 which still has some margin which
will end up above the 32 bit number. Further more the __fls
optimization automatically casts the inputs to unsigned long (32 bit
on 32 bit machines), so that also makes it out of option.
> Also, I expect that this fact could be exploited to optimize this for
> 32bit archs if one were so inclined.
>
>> +{
>> + u64 b, m, y = 0;
>> +
>> + if (x <= 1)
>> + return x;
>> +
>> + m = 1ULL << (64 - 2);
>> + while (m != 0) {
>> + b = y + m;
>> + y >>= 1;
>> +
>> + if (x >= b) {
>> + x -= b;
>> + y += m;
>> + }
>> + m >>= 2;
>> + }
>> + return y;
>> +}
>
> so yeah, no, please try again after reading the current file.
On 18 December 2017 at 17:24, Joe Perches <[email protected]> wrote:
> On Mon, 2017-12-18 at 16:05 +0100, Crt Mori wrote:
>> There is no option to perform 64bit integer sqrt on 32bit platform.
>> Added stronger typed int_sqrt64 enables the 64bit calculations to
>> be performed on 32bit platforms. Although int_sqrt() is a rough
>> approximation, the same algorithm is used in int_sqrt64() as good
>> enough on 32bit platform.
> []
>> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> []
>> @@ -459,6 +459,7 @@ extern int kernel_text_address(unsigned long addr);
>> extern int func_ptr_is_kernel_text(void *ptr);
>>
>> unsigned long int_sqrt(unsigned long);
>> +u64 int_sqrt64(u64 x);
>>
>> extern void bust_spinlocks(int yes);
>> extern int oops_in_progress; /* If set, an oops, panic(), BUG() or die() is in progress */
>> diff --git a/lib/int_sqrt.c b/lib/int_sqrt.c
> []
>> @@ -36,3 +36,29 @@ unsigned long int_sqrt(unsigned long x)
>> return y;
>> }
>> EXPORT_SYMBOL(int_sqrt);
>> +
>> +/**
>> + * int_sqrt64 - strongly typed int_sqrt function
>> + * @x: 64bit integer of which to calculate the sqrt
>> + */
>> +u64 int_sqrt64(u64 x)
>> +{
>> + u64 b, m, y = 0;
>> +
>> + if (x <= 1)
>> + return x;
>> +
>> + m = 1ULL << (64 - 2);
>> + while (m != 0) {
>> + b = y + m;
>> + y >>= 1;
>> +
>> + if (x >= b) {
>> + x -= b;
>> + y += m;
>> + }
>> + m >>= 2;
>> + }
>> + return y;
>> +}
>> +EXPORT_SYMBOL(int_sqrt64);
>
> int_sqrt is now optimized with __fls
>
> Because u64 is unsigned long on 64 bit arches, this
> could probably be optimized for 64 bit distributions
> with something like:
>
> #if BITS_PER_LONG != 64
> #define int_sqrt64 _int_sqrt64
> u64 _int_sqrt64(u64 x);
> #else
> #define int_sqrt64(x) int_sqrt
> #endif
>
> ...
>
> #if BITS_PER_LONG != 64
> u64 _int_sqrt64(u64 x)
> { etc.. }
> #endif
>
Thanks for the tip. I will respin with this change so that 64 bit
arches use the optimized int_sqrt from above.
On Tue, Dec 19, 2017 at 10:31:18AM +0100, Crt Mori wrote:
> IIO kernel does not have the recent version in, so thanks for heads
IIO?
> up. It does not change much for my function.
The comment says:
* Computes: floor(sqrt(x))
floor(sqrt(2^64-1)) == 2^32-1
On 19 December 2017 at 14:25, Peter Zijlstra <[email protected]> wrote:
> On Tue, Dec 19, 2017 at 10:31:18AM +0100, Crt Mori wrote:
>> IIO kernel does not have the recent version in, so thanks for heads
>
> IIO?
>
Industrial Input / Output Subsytem tree
>> up. It does not change much for my function.
>
> The comment says:
>
> * Computes: floor(sqrt(x))
>
> floor(sqrt(2^64-1)) == 2^32-1
>
OK, so return value should be u32.