Skip to contentSkip to conversation

perf tools: Prevent from BITS_PER_LONG redefinition

PATCH In review Frederic Weisbecker 3 participants 7 messages Dec 18, 2009 – Jan 2, 2010 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf tools: Prevent from BITS_PER_LONG redefinition

The system asm/types.h may have defined BITS_PER_LONG already, depending on the distro.

Check that before defining it from bitops.h wrapper.

Fixes:

In file included from util/include/../../../../include/linux/bitops.h:17,

                 from util/include/linux/bitops.h:8,
                 from ../../lib/hweight.c:2:
util/include/asm/bitops.h:9:1: error: "BITS_PER_LONG" redefined
In file included from util/include/../../../../include/linux/bitops.h:3,
                 from util/include/linux/bitops.h:8,
                 from ../../lib/hweight.c:2:

/usr/include/asm/types.h:32:1: error: this is the location of the previous definition
make: *** [util/hweight.o] Erreur 1

Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Paul Mackerras <[email protected]>
---
 tools/perf/util/include/asm/bitops.h |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
tools/perf/util/include/asm/bitops.h +2 -0
index 58e9817..d945334 100644
@@ -5,8 +5,10 @@
 #include "../../types.h"
 #include <linux/compiler.h>
 
+#ifndef BITS_PER_LONG
 /* CHECKME: Not sure both always match */
 #define BITS_PER_LONG	__WORDSIZE
+#endif
 
 #include "../../../../include/asm-generic/bitops/__fls.h"
 #include "../../../../include/asm-generic/bitops/fls.h"
Signature · 1 line
1.6.2.3
maintainer
#1 lore raw

tools/perf/util/include/asm/bitops.h line 11

Quoted 4 lines · Frederic Weisbecker
tools/perf/util/include/asm/bitops.h +2 -0
+#ifndef BITS_PER_LONG
 /* CHECKME: Not sure both always match */
 #define BITS_PER_LONG  __WORDSIZE
+#endif 

Why use __WORDSIZE if you're not sure?

(8*sizeof(long)) is simple and unambiguous.

#2 lore raw
Quoted 9 lines · Peter Zijlstra

tools/perf/util/include/asm/bitops.h line 11

Quoted 4 lines · Frederic Weisbecker
tools/perf/util/include/asm/bitops.h +2 -0
+#ifndef BITS_PER_LONG
 /* CHECKME: Not sure both always match */
 #define BITS_PER_LONG  __WORDSIZE
+#endif 

Why use __WORDSIZE if you're not sure?

(8*sizeof(long)) is simple and unambiguous.

Yeah but we need it from the CPP level.
We include such code located in kernel headers:

static __always_inline unsigned long __fls(unsigned long word)

{
	int num = BITS_PER_LONG - 1;
#if BITS_PER_LONG == 64
	if (!(word & (~0ul << 32))) {
		num -= 32;
		word <<= 32;
	}
#endif
	if (!(word & (~0ul << (BITS_PER_LONG-16)))) {
		num -= 16;
		word <<= 16;
	}

And sizeof() is not defined :)

#3 lore raw
Quoted 19 lines · Frederic Weisbecker

Yeah but we need it from the CPP level.
We include such code located in kernel headers:

static __always_inline unsigned long __fls(unsigned long word)

{
	int num = BITS_PER_LONG - 1;
#if BITS_PER_LONG == 64
	if (!(word&  (~0ul<<  32))) {
		num -= 32;
		word<<= 32;
	}
#endif
	if (!(word&  (~0ul<<  (BITS_PER_LONG-16)))) {
		num -= 16;
		word<<= 16;
	}
Quoted 2 lines

And sizeof() is not defined :)

You can use if() with a constant expression instead of #if.

J

#4 lore raw
Quoted 20 lines · Jeremy Fitzhardinge
Quoted 19 lines · Frederic Weisbecker

Yeah but we need it from the CPP level.
We include such code located in kernel headers:

static __always_inline unsigned long __fls(unsigned long word)

{
	int num = BITS_PER_LONG - 1;
#if BITS_PER_LONG == 64
	if (!(word&  (~0ul<<  32))) {
		num -= 32;
		word<<= 32;
	}
#endif
	if (!(word&  (~0ul<<  (BITS_PER_LONG-16)))) {
		num -= 16;
		word<<= 16;
	}
Quoted 4 lines
Quoted 2 lines

And sizeof() is not defined :)

You can use if() with a constant expression instead of #if.

I did not write this code. But yes you're right, although I think CPP is more suitable here because fls() can be called from fastpath and this conditional build makes one check less and lesser i-cache footprint.

#5 lore raw
Quoted 9 lines · Frederic Weisbecker
Quoted 2 lines

You can use if() with a constant expression instead of #if.

I did not write this code. But yes you're right, although I think CPP is more suitable here because fls() can be called from fastpath and this conditional build makes one check less and lesser i-cache footprint.

Constant if()s are removed at compile time, so there should be no runtime overhead. Constant if()s are generally preferable to #if because the compile will statically check the other code branch, even if it never gets executed, which helps prevent it from rotting.

J

#6 lore raw
Quoted 11 lines · Jeremy Fitzhardinge
Quoted 9 lines · Frederic Weisbecker
Quoted 2 lines

You can use if() with a constant expression instead of #if.

I did not write this code. But yes you're right, although I think CPP is more suitable here because fls() can be called from fastpath and this conditional build makes one check less and lesser i-cache footprint.

Quoted 4 lines

Constant if()s are removed at compile time, so there should be no runtime overhead. Constant if()s are generally preferable to #if because the compile will statically check the other code branch, even if it never gets executed, which helps prevent it from rotting.

Good point! I did not think about compile time optimizations. I'll try that then. Thanks!