lkml.org 
[lkml]   [2001]   [May]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch] 2.4.4 alpha semaphores optimization
On Fri, May 04, 2001 at 02:13:18PM -0700, Richard Henderson wrote:
> Eh? Would you give me an example that isn't working properly?

Sure.

bar.c:
-----------------
extern void rarely_executed_code(void);

static inline void foo_no_be(void)
{
int ret;
__asm__ __volatile__("nop\n": "=r" (ret));
if (ret < 0)
rarely_executed_code();
}

static inline void foo(void)
{
int ret;
__asm__ __volatile__("unop\n": "=r" (ret));
if (__builtin_expect(ret < 0, 0))
rarely_executed_code();
}

#define foo_macro() ({ \
int ret; \
__asm__ __volatile__("fnop\n": "=r" (ret)); \
if (__builtin_expect(ret < 0, 0)) \
rearly_executed_code(); \
})

void bar(void)
{
foo_no_be();
foo();
foo_macro();
}
---------------
bar.s, compiled with 'gcc -O2 -S bar.c':
--------------
[...]
$bar..ng:
lda $30,-16($30)
stq $26,0($30)
.prologue 1
nop

.align 3 #realign
addl $1,$31,$1
blt $1,$L12 # (ret < 0) predicted to be "false", and gcc
# put the code out of line quite nicely.
$L8:
unop

.align 3 #realign
srl $1,31,$1
blbc $1,$L10 # Oops. The slow path code is in line...
jsr $26,rarely_executed_code
ldgp $29,0($26)
$L10:
fnop

.align 3 #realign
srl $1,31,$1
blbs $1,$L13 # This works.
$L11:
ldq $26,0($30)
nop
lda $30,16($30)
ret $31,($26),1
.align 4
$L13:
jsr $26,rearly_executed_code
ldgp $29,0($26)
br $31,$L11
.align 4
$L12:
jsr $26,rarely_executed_code
ldgp $29,0($26)
br $31,$L8
.end bar
.ident "GCC: (GNU) 3.0 20010430 (prerelease)"
---------------
So one of the questions: can one rely on current branch predictions
algorithms (val < 0, val = 0 false etc.) in the long term?

Ivan.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 12:52    [W:0.078 / U:5.516 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site