[PATCH 0/2] perf tools: Fix next_pow2_l()

[PATCH 0/2] perf tools: Fix next_pow2_l()

[Adrian Hunter]

Hi

My implementation of next_pow2_l() was incorrect.
Here is a fix, after which next_pow2() is unused
so I remove it and rename next_pow2_l() -> next_pow2()


Adrian Hunter (2):
      perf tools: Fix next_pow2_l()
      perf tools: Remove unused next_pow2() and rename next_pow2_l()

 tools/perf/util/evlist.c |  2 +-
 tools/perf/util/util.h   | 22 ++++++++--------------
 2 files changed, 9 insertions(+), 15 deletions(-)


Regards
Adrian

[PATCH 1/2] perf tools: Fix next_pow2_l()

[Adrian Hunter]

My implementation of next_pow2_l() was incorrect.
e.g.

  perf record -m4296015872 uname
  rounding mmap pages size to 17592186044416 bytes (4294967296 pages)
  Invalid argument for --mmap_pages/-m

Notice that the next power-of-2 value 4294967296 is
less than the option value 4296015872.

Change to using  __builtin_clzl() and prevent the
shift being equal to the width of the operand.
Also __builtin_clzl(x) is undefined if x is 0,
so adjust the condition to preclude that
possibility.  Now:

  perf record -m4296015872 uname
  rounding mmap pages size to 35184372088832 bytes (8589934592 pages)
  Invalid argument for --mmap_pages/-m

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---
 tools/perf/util/util.h | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index a1eea3e..ae609fe 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -284,13 +284,14 @@ static inline unsigned next_pow2(unsigned x)
 
 static inline unsigned long next_pow2_l(unsigned long x)
 {
-#if BITS_PER_LONG == 64
-	if (x <= (1UL << 31))
-		return next_pow2(x);
-	return (unsigned long)next_pow2(x >> 32) << 32;
-#else
-	return next_pow2(x);
-#endif
+	int leading_zeros;
+
+	if (x < 2)
+		return 1;
+	leading_zeros = __builtin_clzl(x - 1);
+	if (!leading_zeros)
+		return 0;
+	return 1UL << (BITS_PER_LONG - leading_zeros);
 }
 
 size_t hex_width(u64 v);
-- 
1.7.11.7

Re: [PATCH 1/2] perf tools: Fix next_pow2_l()

[Arnaldo Carvalho de Melo]

Em Fri, Dec 13, 2013 at 10:53:35AM +0200, Adrian Hunter escreveu:

> My implementation of next_pow2_l() was incorrect.  e.g.
 
>   perf record -m4296015872 uname
>   rounding mmap pages size to 17592186044416 bytes (4294967296 pages)
>   Invalid argument for --mmap_pages/-m
 
> Notice that the next power-of-2 value 4294967296 is less than the
> option value 4296015872.
 
> Change to using  __builtin_clzl() and prevent the shift being equal to
> the width of the operand.  Also __builtin_clzl(x) is undefined if x is
> 0, so adjust the condition to preclude that possibility.  Now:

Can we try to look first if there is an implementation for what we need
being used in the kernel sources, in include/ and then try to use, if
possible, exactly like done in kernel code?

For instance, include/linux/log2.h has roundup_pow_of_two() that seems
to be what you need, lemme check...

roundup_pow_of_two(4296015872)=8589934592

So, it yields what we need, I'll try to cook up a patch that makes us
use it, as I did in my dwarves tools in this changeset:

https://git.kernel.org/cgit/devel/pahole/pahole.git/commit/?id=e31fda3063e3c88ca0b93a9fb5e6d6e32478e36b

- Arnaldo
	 
>   perf record -m4296015872 uname
>   rounding mmap pages size to 35184372088832 bytes (8589934592 pages)
>   Invalid argument for --mmap_pages/-m
> 
> Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
> ---
>  tools/perf/util/util.h | 15 ++++++++-------
>  1 file changed, 8 insertions(+), 7 deletions(-)
> 
> diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
> index a1eea3e..ae609fe 100644
> --- a/tools/perf/util/util.h
> +++ b/tools/perf/util/util.h
> @@ -284,13 +284,14 @@ static inline unsigned next_pow2(unsigned x)
>  
>  static inline unsigned long next_pow2_l(unsigned long x)
>  {
> -#if BITS_PER_LONG == 64
> -	if (x <= (1UL << 31))
> -		return next_pow2(x);
> -	return (unsigned long)next_pow2(x >> 32) << 32;
> -#else
> -	return next_pow2(x);
> -#endif
> +	int leading_zeros;
> +
> +	if (x < 2)
> +		return 1;
> +	leading_zeros = __builtin_clzl(x - 1);
> +	if (!leading_zeros)
> +		return 0;
> +	return 1UL << (BITS_PER_LONG - leading_zeros);
>  }
>  
>  size_t hex_width(u64 v);
> -- 
> 1.7.11.7

[PATCH 2/2] perf tools: Remove unused next_pow2() and rename next_pow2_l()

[Adrian Hunter]

The fixed version of 'next_pow2_l()' does not call
'next_pow2()' anymore, so it is unused, so remove
it and rename 'next_pow2_l()' to 'next_pow2()'.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---
 tools/perf/util/evlist.c | 2 +-
 tools/perf/util/util.h   | 9 +--------
 2 files changed, 2 insertions(+), 9 deletions(-)

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 0b31cee..327a1a4 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -736,7 +736,7 @@ static long parse_pages_arg(const char *str, unsigned long min,
 		/* leave number of pages at 0 */
 	} else if (!is_power_of_2(pages)) {
 		/* round pages up to next power of 2 */
-		pages = next_pow2_l(pages);
+		pages = next_pow2(pages);
 		if (!pages)
 			return -EINVAL;
 		pr_info("rounding mmap pages size to %lu bytes (%lu pages)\n",
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index ae609fe..3860d76 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -275,14 +275,7 @@ bool is_power_of_2(unsigned long n)
 	return (n != 0 && ((n & (n - 1)) == 0));
 }
 
-static inline unsigned next_pow2(unsigned x)
-{
-	if (!x)
-		return 1;
-	return 1ULL << (32 - __builtin_clz(x - 1));
-}
-
-static inline unsigned long next_pow2_l(unsigned long x)
+static inline unsigned long next_pow2(unsigned long x)
 {
 	int leading_zeros;
 
-- 
1.7.11.7