linux/lib/find_bit.c

271 lines
7.2 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/* bit search implementation
*
* Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*
* Copyright (C) 2008 IBM Corporation
* 'find_last_bit' is written by Rusty Russell <rusty@rustcorp.com.au>
* (Inspired by David Howell's find_next_bit implementation)
*
lib: find_*_bit reimplementation This patchset does rework to find_bit function family to achieve better performance, and decrease size of text. All rework is done in patch 1. Patches 2 and 3 are about code moving and renaming. It was boot-tested on x86_64 and MIPS (big-endian) machines. Performance tests were ran on userspace with code like this: /* addr[] is filled from /dev/urandom */ start = clock(); while (ret < nbits) ret = find_next_bit(addr, nbits, ret + 1); end = clock(); printf("%ld\t", (unsigned long) end - start); On Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz measurements are: (for find_next_bit, nbits is 8M, for find_first_bit - 80K) find_next_bit: find_first_bit: new current new current 26932 43151 14777 14925 26947 43182 14521 15423 26507 43824 15053 14705 27329 43759 14473 14777 26895 43367 14847 15023 26990 43693 15103 15163 26775 43299 15067 15232 27282 42752 14544 15121 27504 43088 14644 14858 26761 43856 14699 15193 26692 43075 14781 14681 27137 42969 14451 15061 ... ... find_next_bit performance gain is 35-40%; find_first_bit - no measurable difference. On ARM machine, there is arch-specific implementation for find_bit. Thanks a lot to George Spelvin and Rasmus Villemoes for hints and helpful discussions. This patch (of 3): New implementations takes less space in source file (see diffstat) and in object. For me it's 710 vs 453 bytes of text. It also shows better performance. find_last_bit description fixed due to obvious typo. [akpm@linux-foundation.org: include linux/bitmap.h, per Rasmus] Signed-off-by: Yury Norov <yury.norov@gmail.com> Reviewed-by: Rasmus Villemoes <linux@rasmusvillemoes.dk> Reviewed-by: George Spelvin <linux@horizon.com> Cc: Alexey Klimov <klimov.linux@gmail.com> Cc: David S. Miller <davem@davemloft.net> Cc: Daniel Borkmann <dborkman@redhat.com> Cc: Hannes Frederic Sowa <hannes@stressinduktion.org> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Mark Salter <msalter@redhat.com> Cc: AKASHI Takahiro <takahiro.akashi@linaro.org> Cc: Thomas Graf <tgraf@suug.ch> Cc: Valentin Rothberg <valentinrothberg@gmail.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-17 03:43:13 +08:00
* Rewritten by Yury Norov <yury.norov@gmail.com> to decrease
* size and improve performance, 2015.
*/
#include <linux/bitops.h>
#include <linux/bitmap.h>
#include <linux/export.h>
#include <linux/math.h>
#include <linux/minmax.h>
#include <linux/swab.h>
/*
* Common helper for find_bit() function family
* @FETCH: The expression that fetches and pre-processes each word of bitmap(s)
* @MUNGE: The expression that post-processes a word containing found bit (may be empty)
* @size: The bitmap size in bits
*/
#define FIND_FIRST_BIT(FETCH, MUNGE, size) \
({ \
unsigned long idx, val, sz = (size); \
\
for (idx = 0; idx * BITS_PER_LONG < sz; idx++) { \
val = (FETCH); \
if (val) { \
sz = min(idx * BITS_PER_LONG + __ffs(MUNGE(val)), sz); \
break; \
} \
} \
\
sz; \
})
/*
lib/find_bit: optimize find_next_bit() functions Over the past couple years, the function _find_next_bit() was extended with parameters that modify its behavior to implement and- zero- and le- flavors. The parameters are passed at compile time, but current design prevents a compiler from optimizing out the conditionals. As find_next_bit() API grows, I expect that more parameters will be added. Current design would require more conditional code in _find_next_bit(), which would bloat the helper even more and make it barely readable. This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds a set of wrappers, so that the compile-time optimizations become possible. The common logic is moved to the new macro, and all flavors may be generated by providing a FETCH macro parameter, like in this example: #define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ... find_next_xornot_and_bit(addr1, addr2, addr3, size, start) { return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx], /* nop */, size, start); } The FETCH may be of any complexity, as soon as it only refers the bitmap(s) and an iterator idx. MUNGE is here to support _le code generation for BE builds. May be empty. I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top of 6.0-rc2 + this series. The results for kvm/x86_64 are: v6.0-rc2 Optimized Difference Z-score Random dense bitmap ns ns ns % find_next_bit: 787735 670546 117189 14.9 3.97 find_next_zero_bit: 777492 664208 113284 14.6 10.51 find_last_bit: 830925 687573 143352 17.3 2.35 find_first_bit: 3874366 3306635 567731 14.7 1.84 find_first_and_bit: 40677125 37739887 2937238 7.2 1.36 find_next_and_bit: 347865 304456 43409 12.5 1.35 Random sparse bitmap find_next_bit: 19816 14021 5795 29.2 6.10 find_next_zero_bit: 1318901 1223794 95107 7.2 1.41 find_last_bit: 14573 13514 1059 7.3 6.92 find_first_bit: 1313321 1249024 64297 4.9 1.53 find_first_and_bit: 8921 8098 823 9.2 4.56 find_next_and_bit: 9796 7176 2620 26.7 5.39 Where the statistics is significant (z-score > 3), the improvement is ~15%. According to the bloat-o-meter, the Image size is 10-11K less: x86_64/defconfig: add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177) arm64/defconfig: add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Yury Norov <yury.norov@gmail.com>
2022-09-15 10:07:29 +08:00
* Common helper for find_next_bit() function family
* @FETCH: The expression that fetches and pre-processes each word of bitmap(s)
* @MUNGE: The expression that post-processes a word containing found bit (may be empty)
* @size: The bitmap size in bits
* @start: The bitnumber to start searching at
*/
lib/find_bit: optimize find_next_bit() functions Over the past couple years, the function _find_next_bit() was extended with parameters that modify its behavior to implement and- zero- and le- flavors. The parameters are passed at compile time, but current design prevents a compiler from optimizing out the conditionals. As find_next_bit() API grows, I expect that more parameters will be added. Current design would require more conditional code in _find_next_bit(), which would bloat the helper even more and make it barely readable. This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds a set of wrappers, so that the compile-time optimizations become possible. The common logic is moved to the new macro, and all flavors may be generated by providing a FETCH macro parameter, like in this example: #define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ... find_next_xornot_and_bit(addr1, addr2, addr3, size, start) { return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx], /* nop */, size, start); } The FETCH may be of any complexity, as soon as it only refers the bitmap(s) and an iterator idx. MUNGE is here to support _le code generation for BE builds. May be empty. I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top of 6.0-rc2 + this series. The results for kvm/x86_64 are: v6.0-rc2 Optimized Difference Z-score Random dense bitmap ns ns ns % find_next_bit: 787735 670546 117189 14.9 3.97 find_next_zero_bit: 777492 664208 113284 14.6 10.51 find_last_bit: 830925 687573 143352 17.3 2.35 find_first_bit: 3874366 3306635 567731 14.7 1.84 find_first_and_bit: 40677125 37739887 2937238 7.2 1.36 find_next_and_bit: 347865 304456 43409 12.5 1.35 Random sparse bitmap find_next_bit: 19816 14021 5795 29.2 6.10 find_next_zero_bit: 1318901 1223794 95107 7.2 1.41 find_last_bit: 14573 13514 1059 7.3 6.92 find_first_bit: 1313321 1249024 64297 4.9 1.53 find_first_and_bit: 8921 8098 823 9.2 4.56 find_next_and_bit: 9796 7176 2620 26.7 5.39 Where the statistics is significant (z-score > 3), the improvement is ~15%. According to the bloat-o-meter, the Image size is 10-11K less: x86_64/defconfig: add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177) arm64/defconfig: add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Yury Norov <yury.norov@gmail.com>
2022-09-15 10:07:29 +08:00
#define FIND_NEXT_BIT(FETCH, MUNGE, size, start) \
({ \
unsigned long mask, idx, tmp, sz = (size), __start = (start); \
\
if (unlikely(__start >= sz)) \
goto out; \
\
mask = MUNGE(BITMAP_FIRST_WORD_MASK(__start)); \
idx = __start / BITS_PER_LONG; \
\
for (tmp = (FETCH) & mask; !tmp; tmp = (FETCH)) { \
if ((idx + 1) * BITS_PER_LONG >= sz) \
goto out; \
idx++; \
} \
\
sz = min(idx * BITS_PER_LONG + __ffs(MUNGE(tmp)), sz); \
out: \
sz; \
})
lib: optimize cpumask_next_and() We've measured that we spend ~0.6% of sys cpu time in cpumask_next_and(). It's essentially a joined iteration in search for a non-zero bit, which is currently implemented as a lookup join (find a nonzero bit on the lhs, lookup the rhs to see if it's set there). Implement a direct join (find a nonzero bit on the incrementally built join). Also add generic bitmap benchmarks in the new `test_find_bit` module for new function (see `find_next_and_bit` in [2] and [3] below). For cpumask_next_and, direct benchmarking shows that it's 1.17x to 14x faster with a geometric mean of 2.1 on 32 CPUs [1]. No impact on memory usage. Note that on Arm, the new pure-C implementation still outperforms the old one that uses a mix of C and asm (`find_next_bit`) [3]. [1] Approximate benchmark code: ``` unsigned long src1p[nr_cpumask_longs] = {pattern1}; unsigned long src2p[nr_cpumask_longs] = {pattern2}; for (/*a bunch of repetitions*/) { for (int n = -1; n <= nr_cpu_ids; ++n) { asm volatile("" : "+rm"(src1p)); // prevent any optimization asm volatile("" : "+rm"(src2p)); unsigned long result = cpumask_next_and(n, src1p, src2p); asm volatile("" : "+rm"(result)); } } ``` Results: pattern1 pattern2 time_before/time_after 0x0000ffff 0x0000ffff 1.65 0x0000ffff 0x00005555 2.24 0x0000ffff 0x00001111 2.94 0x0000ffff 0x00000000 14.0 0x00005555 0x0000ffff 1.67 0x00005555 0x00005555 1.71 0x00005555 0x00001111 1.90 0x00005555 0x00000000 6.58 0x00001111 0x0000ffff 1.46 0x00001111 0x00005555 1.49 0x00001111 0x00001111 1.45 0x00001111 0x00000000 3.10 0x00000000 0x0000ffff 1.18 0x00000000 0x00005555 1.18 0x00000000 0x00001111 1.17 0x00000000 0x00000000 1.25 ----------------------------- geo.mean 2.06 [2] test_find_next_bit, X86 (skylake) [ 3913.477422] Start testing find_bit() with random-filled bitmap [ 3913.477847] find_next_bit: 160868 cycles, 16484 iterations [ 3913.477933] find_next_zero_bit: 169542 cycles, 16285 iterations [ 3913.478036] find_last_bit: 201638 cycles, 16483 iterations [ 3913.480214] find_first_bit: 4353244 cycles, 16484 iterations [ 3913.480216] Start testing find_next_and_bit() with random-filled bitmap [ 3913.481074] find_next_and_bit: 89604 cycles, 8216 iterations [ 3913.481075] Start testing find_bit() with sparse bitmap [ 3913.481078] find_next_bit: 2536 cycles, 66 iterations [ 3913.481252] find_next_zero_bit: 344404 cycles, 32703 iterations [ 3913.481255] find_last_bit: 2006 cycles, 66 iterations [ 3913.481265] find_first_bit: 17488 cycles, 66 iterations [ 3913.481266] Start testing find_next_and_bit() with sparse bitmap [ 3913.481272] find_next_and_bit: 764 cycles, 1 iterations [3] test_find_next_bit, arm (v7 odroid XU3). [ 267.206928] Start testing find_bit() with random-filled bitmap [ 267.214752] find_next_bit: 4474 cycles, 16419 iterations [ 267.221850] find_next_zero_bit: 5976 cycles, 16350 iterations [ 267.229294] find_last_bit: 4209 cycles, 16419 iterations [ 267.279131] find_first_bit: 1032991 cycles, 16420 iterations [ 267.286265] Start testing find_next_and_bit() with random-filled bitmap [ 267.302386] find_next_and_bit: 2290 cycles, 8140 iterations [ 267.309422] Start testing find_bit() with sparse bitmap [ 267.316054] find_next_bit: 191 cycles, 66 iterations [ 267.322726] find_next_zero_bit: 8758 cycles, 32703 iterations [ 267.329803] find_last_bit: 84 cycles, 66 iterations [ 267.336169] find_first_bit: 4118 cycles, 66 iterations [ 267.342627] Start testing find_next_and_bit() with sparse bitmap [ 267.356919] find_next_and_bit: 91 cycles, 1 iterations [courbet@google.com: v6] Link: http://lkml.kernel.org/r/20171129095715.23430-1-courbet@google.com [geert@linux-m68k.org: m68k/bitops: always include <asm-generic/bitops/find.h>] Link: http://lkml.kernel.org/r/1512556816-28627-1-git-send-email-geert@linux-m68k.org Link: http://lkml.kernel.org/r/20171128131334.23491-1-courbet@google.com Signed-off-by: Clement Courbet <courbet@google.com> Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Yury Norov <ynorov@caviumnetworks.com> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-02-07 07:38:34 +08:00
#define FIND_NTH_BIT(FETCH, size, num) \
({ \
unsigned long sz = (size), nr = (num), idx, w, tmp; \
\
for (idx = 0; (idx + 1) * BITS_PER_LONG <= sz; idx++) { \
if (idx * BITS_PER_LONG + nr >= sz) \
goto out; \
\
tmp = (FETCH); \
w = hweight_long(tmp); \
if (w > nr) \
goto found; \
\
nr -= w; \
} \
\
if (sz % BITS_PER_LONG) \
tmp = (FETCH) & BITMAP_LAST_WORD_MASK(sz); \
found: \
sz = min(idx * BITS_PER_LONG + fns(tmp, nr), sz); \
out: \
sz; \
})
#ifndef find_first_bit
/*
* Find the first set bit in a memory region.
*/
unsigned long _find_first_bit(const unsigned long *addr, unsigned long size)
{
return FIND_FIRST_BIT(addr[idx], /* nop */, size);
}
EXPORT_SYMBOL(_find_first_bit);
#endif
#ifndef find_first_and_bit
/*
* Find the first set bit in two memory regions.
*/
unsigned long _find_first_and_bit(const unsigned long *addr1,
const unsigned long *addr2,
unsigned long size)
{
return FIND_FIRST_BIT(addr1[idx] & addr2[idx], /* nop */, size);
}
EXPORT_SYMBOL(_find_first_and_bit);
#endif
#ifndef find_first_zero_bit
/*
* Find the first cleared bit in a memory region.
*/
unsigned long _find_first_zero_bit(const unsigned long *addr, unsigned long size)
{
return FIND_FIRST_BIT(~addr[idx], /* nop */, size);
}
EXPORT_SYMBOL(_find_first_zero_bit);
#endif
lib/find_bit: optimize find_next_bit() functions Over the past couple years, the function _find_next_bit() was extended with parameters that modify its behavior to implement and- zero- and le- flavors. The parameters are passed at compile time, but current design prevents a compiler from optimizing out the conditionals. As find_next_bit() API grows, I expect that more parameters will be added. Current design would require more conditional code in _find_next_bit(), which would bloat the helper even more and make it barely readable. This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds a set of wrappers, so that the compile-time optimizations become possible. The common logic is moved to the new macro, and all flavors may be generated by providing a FETCH macro parameter, like in this example: #define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ... find_next_xornot_and_bit(addr1, addr2, addr3, size, start) { return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx], /* nop */, size, start); } The FETCH may be of any complexity, as soon as it only refers the bitmap(s) and an iterator idx. MUNGE is here to support _le code generation for BE builds. May be empty. I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top of 6.0-rc2 + this series. The results for kvm/x86_64 are: v6.0-rc2 Optimized Difference Z-score Random dense bitmap ns ns ns % find_next_bit: 787735 670546 117189 14.9 3.97 find_next_zero_bit: 777492 664208 113284 14.6 10.51 find_last_bit: 830925 687573 143352 17.3 2.35 find_first_bit: 3874366 3306635 567731 14.7 1.84 find_first_and_bit: 40677125 37739887 2937238 7.2 1.36 find_next_and_bit: 347865 304456 43409 12.5 1.35 Random sparse bitmap find_next_bit: 19816 14021 5795 29.2 6.10 find_next_zero_bit: 1318901 1223794 95107 7.2 1.41 find_last_bit: 14573 13514 1059 7.3 6.92 find_first_bit: 1313321 1249024 64297 4.9 1.53 find_first_and_bit: 8921 8098 823 9.2 4.56 find_next_and_bit: 9796 7176 2620 26.7 5.39 Where the statistics is significant (z-score > 3), the improvement is ~15%. According to the bloat-o-meter, the Image size is 10-11K less: x86_64/defconfig: add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177) arm64/defconfig: add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Yury Norov <yury.norov@gmail.com>
2022-09-15 10:07:29 +08:00
#ifndef find_next_bit
unsigned long _find_next_bit(const unsigned long *addr, unsigned long nbits, unsigned long start)
{
return FIND_NEXT_BIT(addr[idx], /* nop */, nbits, start);
}
EXPORT_SYMBOL(_find_next_bit);
#endif
unsigned long __find_nth_bit(const unsigned long *addr, unsigned long size, unsigned long n)
{
return FIND_NTH_BIT(addr[idx], size, n);
}
EXPORT_SYMBOL(__find_nth_bit);
unsigned long __find_nth_and_bit(const unsigned long *addr1, const unsigned long *addr2,
unsigned long size, unsigned long n)
{
return FIND_NTH_BIT(addr1[idx] & addr2[idx], size, n);
}
EXPORT_SYMBOL(__find_nth_and_bit);
unsigned long __find_nth_andnot_bit(const unsigned long *addr1, const unsigned long *addr2,
unsigned long size, unsigned long n)
{
return FIND_NTH_BIT(addr1[idx] & ~addr2[idx], size, n);
}
EXPORT_SYMBOL(__find_nth_andnot_bit);
unsigned long __find_nth_and_andnot_bit(const unsigned long *addr1,
const unsigned long *addr2,
const unsigned long *addr3,
unsigned long size, unsigned long n)
{
return FIND_NTH_BIT(addr1[idx] & addr2[idx] & ~addr3[idx], size, n);
}
EXPORT_SYMBOL(__find_nth_and_andnot_bit);
lib/find_bit: optimize find_next_bit() functions Over the past couple years, the function _find_next_bit() was extended with parameters that modify its behavior to implement and- zero- and le- flavors. The parameters are passed at compile time, but current design prevents a compiler from optimizing out the conditionals. As find_next_bit() API grows, I expect that more parameters will be added. Current design would require more conditional code in _find_next_bit(), which would bloat the helper even more and make it barely readable. This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds a set of wrappers, so that the compile-time optimizations become possible. The common logic is moved to the new macro, and all flavors may be generated by providing a FETCH macro parameter, like in this example: #define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ... find_next_xornot_and_bit(addr1, addr2, addr3, size, start) { return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx], /* nop */, size, start); } The FETCH may be of any complexity, as soon as it only refers the bitmap(s) and an iterator idx. MUNGE is here to support _le code generation for BE builds. May be empty. I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top of 6.0-rc2 + this series. The results for kvm/x86_64 are: v6.0-rc2 Optimized Difference Z-score Random dense bitmap ns ns ns % find_next_bit: 787735 670546 117189 14.9 3.97 find_next_zero_bit: 777492 664208 113284 14.6 10.51 find_last_bit: 830925 687573 143352 17.3 2.35 find_first_bit: 3874366 3306635 567731 14.7 1.84 find_first_and_bit: 40677125 37739887 2937238 7.2 1.36 find_next_and_bit: 347865 304456 43409 12.5 1.35 Random sparse bitmap find_next_bit: 19816 14021 5795 29.2 6.10 find_next_zero_bit: 1318901 1223794 95107 7.2 1.41 find_last_bit: 14573 13514 1059 7.3 6.92 find_first_bit: 1313321 1249024 64297 4.9 1.53 find_first_and_bit: 8921 8098 823 9.2 4.56 find_next_and_bit: 9796 7176 2620 26.7 5.39 Where the statistics is significant (z-score > 3), the improvement is ~15%. According to the bloat-o-meter, the Image size is 10-11K less: x86_64/defconfig: add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177) arm64/defconfig: add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Yury Norov <yury.norov@gmail.com>
2022-09-15 10:07:29 +08:00
#ifndef find_next_and_bit
unsigned long _find_next_and_bit(const unsigned long *addr1, const unsigned long *addr2,
unsigned long nbits, unsigned long start)
{
return FIND_NEXT_BIT(addr1[idx] & addr2[idx], /* nop */, nbits, start);
}
EXPORT_SYMBOL(_find_next_and_bit);
#endif
#ifndef find_next_andnot_bit
unsigned long _find_next_andnot_bit(const unsigned long *addr1, const unsigned long *addr2,
unsigned long nbits, unsigned long start)
{
return FIND_NEXT_BIT(addr1[idx] & ~addr2[idx], /* nop */, nbits, start);
}
EXPORT_SYMBOL(_find_next_andnot_bit);
#endif
#ifndef find_next_or_bit
unsigned long _find_next_or_bit(const unsigned long *addr1, const unsigned long *addr2,
unsigned long nbits, unsigned long start)
{
return FIND_NEXT_BIT(addr1[idx] | addr2[idx], /* nop */, nbits, start);
}
EXPORT_SYMBOL(_find_next_or_bit);
#endif
lib/find_bit: optimize find_next_bit() functions Over the past couple years, the function _find_next_bit() was extended with parameters that modify its behavior to implement and- zero- and le- flavors. The parameters are passed at compile time, but current design prevents a compiler from optimizing out the conditionals. As find_next_bit() API grows, I expect that more parameters will be added. Current design would require more conditional code in _find_next_bit(), which would bloat the helper even more and make it barely readable. This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds a set of wrappers, so that the compile-time optimizations become possible. The common logic is moved to the new macro, and all flavors may be generated by providing a FETCH macro parameter, like in this example: #define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ... find_next_xornot_and_bit(addr1, addr2, addr3, size, start) { return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx], /* nop */, size, start); } The FETCH may be of any complexity, as soon as it only refers the bitmap(s) and an iterator idx. MUNGE is here to support _le code generation for BE builds. May be empty. I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top of 6.0-rc2 + this series. The results for kvm/x86_64 are: v6.0-rc2 Optimized Difference Z-score Random dense bitmap ns ns ns % find_next_bit: 787735 670546 117189 14.9 3.97 find_next_zero_bit: 777492 664208 113284 14.6 10.51 find_last_bit: 830925 687573 143352 17.3 2.35 find_first_bit: 3874366 3306635 567731 14.7 1.84 find_first_and_bit: 40677125 37739887 2937238 7.2 1.36 find_next_and_bit: 347865 304456 43409 12.5 1.35 Random sparse bitmap find_next_bit: 19816 14021 5795 29.2 6.10 find_next_zero_bit: 1318901 1223794 95107 7.2 1.41 find_last_bit: 14573 13514 1059 7.3 6.92 find_first_bit: 1313321 1249024 64297 4.9 1.53 find_first_and_bit: 8921 8098 823 9.2 4.56 find_next_and_bit: 9796 7176 2620 26.7 5.39 Where the statistics is significant (z-score > 3), the improvement is ~15%. According to the bloat-o-meter, the Image size is 10-11K less: x86_64/defconfig: add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177) arm64/defconfig: add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Yury Norov <yury.norov@gmail.com>
2022-09-15 10:07:29 +08:00
#ifndef find_next_zero_bit
unsigned long _find_next_zero_bit(const unsigned long *addr, unsigned long nbits,
unsigned long start)
{
return FIND_NEXT_BIT(~addr[idx], /* nop */, nbits, start);
}
EXPORT_SYMBOL(_find_next_zero_bit);
#endif
#ifndef find_last_bit
unsigned long _find_last_bit(const unsigned long *addr, unsigned long size)
{
if (size) {
unsigned long val = BITMAP_LAST_WORD_MASK(size);
unsigned long idx = (size-1) / BITS_PER_LONG;
do {
val &= addr[idx];
if (val)
return idx * BITS_PER_LONG + __fls(val);
val = ~0ul;
} while (idx--);
}
return size;
}
EXPORT_SYMBOL(_find_last_bit);
#endif
bitops: introduce the for_each_set_clump8 macro Pach series "Introduce the for_each_set_clump8 macro", v18. While adding GPIO get_multiple/set_multiple callback support for various drivers, I noticed a pattern of looping manifesting that would be useful standardized as a macro. This patchset introduces the for_each_set_clump8 macro and utilizes it in several GPIO drivers. The for_each_set_clump macro8 facilitates a for-loop syntax that iterates over a memory region entire groups of set bits at a time. For example, suppose you would like to iterate over a 32-bit integer 8 bits at a time, skipping over 8-bit groups with no set bit, where XXXXXXXX represents the current 8-bit group: Example: 10111110 00000000 11111111 00110011 First loop: 10111110 00000000 11111111 XXXXXXXX Second loop: 10111110 00000000 XXXXXXXX 00110011 Third loop: XXXXXXXX 00000000 11111111 00110011 Each iteration of the loop returns the next 8-bit group that has at least one set bit. The for_each_set_clump8 macro has four parameters: * start: set to the bit offset of the current clump * clump: set to the current clump value * bits: bitmap to search within * size: bitmap size in number of bits In this version of the patchset, the for_each_set_clump macro has been reimplemented and simplified based on the suggestions provided by Rasmus Villemoes and Andy Shevchenko in the version 4 submission. In particular, the function of the for_each_set_clump macro has been restricted to handle only 8-bit clumps; the drivers that use the for_each_set_clump macro only handle 8-bit ports so a generic for_each_set_clump implementation is not necessary. Thus, a solution for large clumps (i.e. those larger than the width of a bitmap word) can be postponed until a driver appears that actually requires such a generic for_each_set_clump implementation. For what it's worth, a semi-generic for_each_set_clump (i.e. for clumps smaller than the width of a bitmap word) can be implemented by simply replacing the hardcoded '8' and '0xFF' instances with respective variables. I have not yet had a need for such an implementation, and since it falls short of a true generic for_each_set_clump function, I have decided to forgo such an implementation for now. In addition, the bitmap_get_value8 and bitmap_set_value8 functions are introduced to get and set 8-bit values respectively. Their use is based on the behavior suggested in the patchset version 4 review. This patch (of 14): This macro iterates for each 8-bit group of bits (clump) with set bits, within a bitmap memory region. For each iteration, "start" is set to the bit offset of the found clump, while the respective clump value is stored to the location pointed by "clump". Additionally, the bitmap_get_value8 and bitmap_set_value8 functions are introduced to respectively get and set an 8-bit value in a bitmap memory region. [gustavo@embeddedor.com: fix potential sign-extension overflow] Link: http://lkml.kernel.org/r/20191015184657.GA26541@embeddedor [akpm@linux-foundation.org: s/ULL/UL/, per Joe] [vilhelm.gray@gmail.com: add for_each_set_clump8 documentation] Link: http://lkml.kernel.org/r/20191016161825.301082-1-vilhelm.gray@gmail.com Link: http://lkml.kernel.org/r/893c3b4f03266c9496137cc98ac2b1bd27f92c73.1570641097.git.vilhelm.gray@gmail.com Signed-off-by: William Breathitt Gray <vilhelm.gray@gmail.com> Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com> Suggested-by: Andy Shevchenko <andy.shevchenko@gmail.com> Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk> Suggested-by: Lukas Wunner <lukas@wunner.de> Tested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Linus Walleij <linus.walleij@linaro.org> Cc: Bartosz Golaszewski <bgolaszewski@baylibre.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Phil Reid <preid@electromag.com.au> Cc: Geert Uytterhoeven <geert+renesas@glider.be> Cc: Mathias Duckeck <m.duckeck@kunbus.de> Cc: Morten Hein Tiljeset <morten.tiljeset@prevas.dk> Cc: Sean Nyekjaer <sean.nyekjaer@prevas.dk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-05 08:50:57 +08:00
unsigned long find_next_clump8(unsigned long *clump, const unsigned long *addr,
unsigned long size, unsigned long offset)
{
offset = find_next_bit(addr, size, offset);
if (offset == size)
return size;
offset = round_down(offset, 8);
*clump = bitmap_get_value8(addr, offset);
return offset;
}
EXPORT_SYMBOL(find_next_clump8);
#ifdef __BIG_ENDIAN
#ifndef find_first_zero_bit_le
/*
* Find the first cleared bit in an LE memory region.
*/
unsigned long _find_first_zero_bit_le(const unsigned long *addr, unsigned long size)
{
return FIND_FIRST_BIT(~addr[idx], swab, size);
}
EXPORT_SYMBOL(_find_first_zero_bit_le);
#endif
lib/find_bit: optimize find_next_bit() functions Over the past couple years, the function _find_next_bit() was extended with parameters that modify its behavior to implement and- zero- and le- flavors. The parameters are passed at compile time, but current design prevents a compiler from optimizing out the conditionals. As find_next_bit() API grows, I expect that more parameters will be added. Current design would require more conditional code in _find_next_bit(), which would bloat the helper even more and make it barely readable. This patch replaces _find_next_bit() with a macro FIND_NEXT_BIT, and adds a set of wrappers, so that the compile-time optimizations become possible. The common logic is moved to the new macro, and all flavors may be generated by providing a FETCH macro parameter, like in this example: #define FIND_NEXT_BIT(FETCH, MUNGE, size, start) ... find_next_xornot_and_bit(addr1, addr2, addr3, size, start) { return FIND_NEXT_BIT(addr1[idx] ^ ~addr2[idx] & addr3[idx], /* nop */, size, start); } The FETCH may be of any complexity, as soon as it only refers the bitmap(s) and an iterator idx. MUNGE is here to support _le code generation for BE builds. May be empty. I ran find_bit_benchmark 16 times on top of 6.0-rc2 and 16 times on top of 6.0-rc2 + this series. The results for kvm/x86_64 are: v6.0-rc2 Optimized Difference Z-score Random dense bitmap ns ns ns % find_next_bit: 787735 670546 117189 14.9 3.97 find_next_zero_bit: 777492 664208 113284 14.6 10.51 find_last_bit: 830925 687573 143352 17.3 2.35 find_first_bit: 3874366 3306635 567731 14.7 1.84 find_first_and_bit: 40677125 37739887 2937238 7.2 1.36 find_next_and_bit: 347865 304456 43409 12.5 1.35 Random sparse bitmap find_next_bit: 19816 14021 5795 29.2 6.10 find_next_zero_bit: 1318901 1223794 95107 7.2 1.41 find_last_bit: 14573 13514 1059 7.3 6.92 find_first_bit: 1313321 1249024 64297 4.9 1.53 find_first_and_bit: 8921 8098 823 9.2 4.56 find_next_and_bit: 9796 7176 2620 26.7 5.39 Where the statistics is significant (z-score > 3), the improvement is ~15%. According to the bloat-o-meter, the Image size is 10-11K less: x86_64/defconfig: add/remove: 32/14 grow/shrink: 61/782 up/down: 6344/-16521 (-10177) arm64/defconfig: add/remove: 3/2 grow/shrink: 50/714 up/down: 608/-11556 (-10948) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Yury Norov <yury.norov@gmail.com>
2022-09-15 10:07:29 +08:00
#ifndef find_next_zero_bit_le
unsigned long _find_next_zero_bit_le(const unsigned long *addr,
unsigned long size, unsigned long offset)
{
return FIND_NEXT_BIT(~addr[idx], swab, size, offset);
}
EXPORT_SYMBOL(_find_next_zero_bit_le);
#endif
#ifndef find_next_bit_le
unsigned long _find_next_bit_le(const unsigned long *addr,
unsigned long size, unsigned long offset)
{
return FIND_NEXT_BIT(addr[idx], swab, size, offset);
}
EXPORT_SYMBOL(_find_next_bit_le);
#endif
#endif /* __BIG_ENDIAN */