On Fri, Apr 26, 2024 at 12:48:48PM -0700, Yury Norov wrote: > On Fri, Apr 26, 2024 at 12:08 PM Andrew Morton > <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > > > The patch titled > > Subject: bitops: optimize fns() for improved performance > > has been added to the -mm mm-nonmm-unstable branch. Its filename is > > bitops-optimize-fns-for-improved-performance.patch > > > > This patch will shortly appear at > > https://git.kernel.org/pub/scm/linux/kernel/git/akpm/25-new.git/tree/patches/bitops-optimize-fns-for-improved-performance.patch > > > > This patch will later appear in the mm-nonmm-unstable branch at > > git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm > > > > Before you just go and hit "reply", please: > > a) Consider who else should be cc'ed > > b) Prefer to cc a suitable mailing list as well > > c) Ideally: find the original patch on the mailing list and do a > > reply-to-all to that, adding suitable additional cc's > > > > *** Remember to use Documentation/process/submit-checklist.rst when testing your code *** > > > > The -mm tree is included into linux-next via the mm-everything > > branch at git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm > > and is updated there every 2-3 working days > > > > ------------------------------------------------------ > > From: Kuan-Wei Chiu <visitorckw@xxxxxxxxx> > > Subject: bitops: optimize fns() for improved performance > > Date: Fri, 26 Apr 2024 11:51:52 +0800 > > > > The current fns() repeatedly uses __ffs() to find the index of the least > > significant bit and then clears the corresponding bit using __clear_bit(). > > The method for clearing the least significant bit can be optimized by > > using word &= word - 1 instead. > > > > Typically, the execution time of one __ffs() plus one __clear_bit() is > > longer than that of a bitwise AND operation and a subtraction. To improve > > performance, the loop for clearing the least significant bit has been > > replaced with word &= word - 1, followed by a single __ffs() operation to > > obtain the answer. This change reduces the number of __ffs() iterations > > from n to just one, enhancing overall performance. > > > > The following microbenchmark data, conducted on my x86-64 machine, shows > > the execution time (in microseconds) required for 1000000 test data > > generated by get_random_u64() and executed by fns() under different values > > of n: > > > > +-----+---------------+---------------+ > > | n | time_old | time_new | > > +-----+---------------+---------------+ > > | 0 | 29194 | 25878 | > > | 1 | 25510 | 25497 | > > | 2 | 27836 | 25721 | > > | 3 | 30140 | 25673 | > > | 4 | 32569 | 25426 | > > | 5 | 34792 | 25690 | > > | 6 | 37117 | 25651 | > > | 7 | 39742 | 25383 | > > | 8 | 42360 | 25657 | > > | 9 | 44672 | 25897 | > > | 10 | 47237 | 25819 | > > | 11 | 49884 | 26530 | > > | 12 | 51864 | 26647 | > > | 13 | 54265 | 28915 | > > | 14 | 56440 | 28373 | > > | 15 | 58839 | 28616 | > > | 16 | 62383 | 29128 | > > | 17 | 64257 | 30041 | > > | 18 | 66805 | 29773 | > > | 19 | 69368 | 33203 | > > | 20 | 72942 | 33688 | > > | 21 | 77006 | 34518 | > > | 22 | 80926 | 34298 | > > | 23 | 85723 | 35586 | > > | 24 | 90324 | 36376 | > > | 25 | 95992 | 37465 | > > | 26 | 101101 | 37599 | > > | 27 | 106520 | 37466 | > > | 28 | 113287 | 38163 | > > | 29 | 120552 | 38810 | > > | 30 | 128040 | 39373 | > > | 31 | 135624 | 40500 | > > | 32 | 142580 | 40343 | > > | 33 | 148915 | 40460 | > > | 34 | 154005 | 41294 | > > | 35 | 157996 | 41730 | > > | 36 | 160806 | 41523 | > > | 37 | 162975 | 42088 | > > | 38 | 163426 | 41530 | > > | 39 | 164872 | 41789 | > > | 40 | 164477 | 42505 | > > | 41 | 164758 | 41879 | > > | 42 | 164182 | 41415 | > > | 43 | 164842 | 42119 | > > | 44 | 164881 | 42297 | > > | 45 | 164870 | 42145 | > > | 46 | 164673 | 42066 | > > | 47 | 164616 | 42051 | > > | 48 | 165055 | 41902 | > > | 49 | 164847 | 41862 | > > | 50 | 165171 | 41960 | > > | 51 | 164851 | 42089 | > > | 52 | 164763 | 41717 | > > | 53 | 164635 | 42154 | > > | 54 | 164757 | 41983 | > > | 55 | 165095 | 41419 | > > | 56 | 164641 | 42381 | > > | 57 | 164601 | 41654 | > > | 58 | 164864 | 41834 | > > | 59 | 164594 | 41920 | > > | 60 | 165207 | 42020 | > > | 61 | 165056 | 41185 | > > | 62 | 165160 | 41722 | > > | 63 | 164923 | 41702 | > > | 64 | 164777 | 41880 | > > +-----+---------------+---------------+ > > Hi Kuan-Wei, > > I didn't receive the original email for some reason... > We've got a performance test for the function in find_bit_benchmark. > Can you print before/after here? > > Thanks, > Yury > Hi Yury, Here are the benchmark results: Before: Start testing find_bit() with random-filled bitmap [ 0.299085] fbcon: Taking over console [ 0.299820] find_next_bit: 606286 ns, 164169 iterations [ 0.300463] find_next_zero_bit: 641072 ns, 163512 iterations [ 0.300996] find_last_bit: 531027 ns, 164169 iterations [ 0.305233] find_nth_bit: 4235859 ns, 16454 iterations [ 0.306434] find_first_bit: 1199357 ns, 16455 iterations [ 0.321616] find_first_and_bit: 15179667 ns, 32869 iterations [ 0.321917] find_next_and_bit: 298836 ns, 73875 iterations [ 0.321918] Start testing find_bit() with sparse bitmap [ 0.321953] find_next_bit: 7931 ns, 656 iterations [ 0.323201] find_next_zero_bit: 1246980 ns, 327025 iterations [ 0.323210] find_last_bit: 8000 ns, 656 iterations [ 0.324427] find_nth_bit: 1213161 ns, 655 iterations [ 0.324813] find_first_bit: 384747 ns, 656 iterations [ 0.324817] find_first_and_bit: 2220 ns, 1 iterations [ 0.324820] find_next_and_bit: 1831 ns, 1 iterations After: Start testing find_bit() with random-filled bitmap [ 0.305081] fbcon: Taking over console [ 0.306126] find_next_bit: 854517 ns, 163960 iterations [ 0.307041] find_next_zero_bit: 911725 ns, 163721 iterations [ 0.307711] find_last_bit: 668261 ns, 163960 iterations [ 0.311160] find_nth_bit: 3447530 ns, 16372 iterations [ 0.312358] find_first_bit: 1196633 ns, 16373 iterations [ 0.327191] find_first_and_bit: 14830129 ns, 32951 iterations [ 0.327503] find_next_and_bit: 310560 ns, 73719 iterations [ 0.327504] Start testing find_bit() with sparse bitmap [ 0.327539] find_next_bit: 7633 ns, 656 iterations [ 0.328787] find_next_zero_bit: 1247398 ns, 327025 iterations [ 0.328797] find_last_bit: 8425 ns, 656 iterations [ 0.330034] find_nth_bit: 1234044 ns, 655 iterations [ 0.330428] find_first_bit: 392086 ns, 656 iterations [ 0.330431] find_first_and_bit: 1980 ns, 1 iterations [ 0.330434] find_next_and_bit: 1831 ns, 1 iterations Some benchmarks seem to have worsened after applying this patch. However, unless I'm mistaken, the fns() changes should only affect the results of find_nth_bit, while the others are just random fluctuations. Should I include the above benchmark data in the commit message and send a v2 patch? Additionally, I apologize for you not receiving the email. I received the following "Message not delivered" email, but I'm unsure if it's related and what caused the error: Date: Sat, 27 Apr 2024 04:29:04 +0000 (UTC) From: do-not-reply@xxxxxxxxxxxxxxx To: visitorckw@xxxxxxxxx Subject: Undelivered Mail This is an automated message from mail service of vivek.yagnik@xxxxxxxxxxxxxxx ⚠ Message not delivered ------------------ Message details ------------------ From: visitorckw@xxxxxxxxx To: vivek.yagnik@xxxxxxxxxxxxxxx Sent: 2024-04-27T04:29:03.000Z Subject: [PATCH] bitops: Optimize fns() for improved performance Failure reason: <vivek.yagnik@xxxxxxxxxxxxxxx>: host sophosemail-com.mail.protection.outlook.com[52.101.144.3] +said: 451 4.4.4 Mail received as unauthenticated, incoming to a recipient domain configured in a hosted tenant +which has no mail-enabled subscriptions. ATTR5 [MA1PEPF000072B2.INDPRD01.PROD.OUTLOOK.COM 2024-04-27T04:29:03.836Z +08DC631634A0BBEB] (in reply to end of DATA command) Regards, Kuan-Wei > > Link: https://lkml.kernel.org/r/20240426035152.956702-1-visitorckw@xxxxxxxxx > > Signed-off-by: Kuan-Wei Chiu <visitorckw@xxxxxxxxx> > > Cc: Ching-Chun (Jim) Huang <jserv@xxxxxxxxxxxxxxxx> > > Cc: Yury Norov <yury.norov@xxxxxxxxx> > > Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> > > --- > > > > include/linux/bitops.h | 12 ++++-------- > > 1 file changed, 4 insertions(+), 8 deletions(-) > > > > --- a/include/linux/bitops.h~bitops-optimize-fns-for-improved-performance > > +++ a/include/linux/bitops.h > > @@ -254,16 +254,12 @@ static inline unsigned long __ffs64(u64 > > */ > > static inline unsigned long fns(unsigned long word, unsigned int n) > > { > > - unsigned int bit; > > + unsigned int i; > > > > - while (word) { > > - bit = __ffs(word); > > - if (n-- == 0) > > - return bit; > > - __clear_bit(bit, &word); > > - } > > + for (i = 0; word && i < n; i++) > > + word &= word - 1; > > > > - return BITS_PER_LONG; > > + return word ? __ffs(word) : BITS_PER_LONG; > > } > > > > /** > > _ > > > > Patches currently in -mm which might be from visitorckw@xxxxxxxxx are > > > > bitops-optimize-fns-for-improved-performance.patch > >
