Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
[PATCH] speed up on find_first_bit for i386 (let compiler do the work)
Avoid using "rep scas", just let the compiler select a sequence of
regular instructions.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
authored by
Steven Rostedt
and committed by
Linus Torvalds
cd85c8b4
79a88102
+24
-30
+24
-30
include/asm-i386/bitops.h
+24
-30
include/asm-i386/bitops.h
···
311
311
int find_next_zero_bit(const unsigned long *addr, int size, int offset);
312
312
313
313
/**
314
+
* __ffs - find first bit in word.
315
+
* @word: The word to search
316
+
*
317
+
* Undefined if no bit exists, so code should check against 0 first.
318
+
*/
319
+
static inline unsigned long __ffs(unsigned long word)
320
+
{
321
+
__asm__("bsfl %1,%0"
322
+
:"=r" (word)
323
+
:"rm" (word));
324
+
return word;
325
+
}
326
+
327
+
/**
314
328
* find_first_bit - find the first set bit in a memory region
315
329
* @addr: The address to start the search at
316
330
* @size: The maximum size to search
···
334
320
*/
335
321
static inline int find_first_bit(const unsigned long *addr, unsigned size)
336
322
{
337
-
int d0, d1;
338
-
int res;
339
-
340
-
/* This looks at memory. Mark it volatile to tell gcc not to move it around */
341
-
__asm__ __volatile__(
342
-
"xorl %%eax,%%eax\n\t"
343
-
"repe; scasl\n\t"
344
-
"jz 1f\n\t"
345
-
"leal -4(%%edi),%%edi\n\t"
346
-
"bsfl (%%edi),%%eax\n"
347
-
"1:\tsubl %%ebx,%%edi\n\t"
348
-
"shll $3,%%edi\n\t"
349
-
"addl %%edi,%%eax"
350
-
:"=a" (res), "=&c" (d0), "=&D" (d1)
351
-
:"1" ((size + 31) >> 5), "2" (addr), "b" (addr) : "memory");
352
-
return res;
323
+
int x = 0;
324
+
do {
325
+
if (*addr)
326
+
return __ffs(*addr) + x;
327
+
addr++;
328
+
if (x >= size)
329
+
break;
330
+
x += (sizeof(*addr)<<3);
331
+
} while (1);
332
+
return x;
353
333
}
354
334
355
335
/**
···
365
357
__asm__("bsfl %1,%0"
366
358
:"=r" (word)
367
359
:"r" (~word));
368
-
return word;
369
-
}
370
-
371
-
/**
372
-
* __ffs - find first bit in word.
373
-
* @word: The word to search
374
-
*
375
-
* Undefined if no bit exists, so code should check against 0 first.
376
-
*/
377
-
static inline unsigned long __ffs(unsigned long word)
378
-
{
379
-
__asm__("bsfl %1,%0"
380
-
:"=r" (word)
381
-
:"rm" (word));
382
360
return word;
383
361
}
384
362