Replace the bit by bit algorithm with one that generates 16 bits per
iteration on 32bit architectures and 32 bits on 64bit ones.
On my zen 5 this reduces the time for the tests (using the generic code)
from ~3350ns to ~1000ns.
Running the 32bit algorithm on 64bit x86 takes ~1500ns. It'll be slightly
slower on a real 32bit system, mostly due to register pressure.
The savings for 32bit x86 are much higher (tested in userspace). The
worst case (lots of bits in the quotient) drops from ~900 clocks to ~130
(pretty much independant of the arguments). Other 32bit architectures may
see better savings.
It is possibly to optimise for divisors that span less than
__LONG_WIDTH__/2 bits. However I suspect they don't happen that often and
it doesn't remove any slow cpu divide instructions which dominate the
result.
Typical improvements for 64bit random divides:
old new
sandy bridge: 470 150
haswell: 400 144
piledriver: 960 467 I think rdpmc is very slow.
zen5: 244 80
(Timing is 'rdpmc; mul_div(); rdpmc' with the multiply depending on the
first rdpmc and the second rdpmc depending on the quotient.)
Object code (64bit x86 test program): old 0x173 new 0x141.
Link: https://lkml.kernel.org/r/20251105201035.64043-9-david.laight.linux@gmail.com
Signed-off-by: David Laight <david.laight.linux@gmail.com>
Reviewed-by: Nicolas Pitre <npitre@baylibre.com>
Cc: Biju Das <biju.das.jz@bp.renesas.com>
Cc: Borislav Betkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleinxer <tglx@linutronix.de>
Cc: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
gcc generates horrid code for both ((u64)u32_a * u32_b) and (u64_a +
u32_b). As well as the extra instructions it can generate a lot of spills
to stack (including spills of constant zeros and even multiplies by
constant zero).
mul_u32_u32() already exists to optimise the multiply. Add a similar
add_u64_32() for the addition. Disable both for clang - it generates
better code without them.
Move the 64x64 => 128 multiply into a static inline helper function for
code clarity. No need for the a/b_hi/lo variables, the implicit casts on
the function calls do the work for us. Should have minimal effect on the
generated code.
Use mul_u32_u32() and add_u64_u32() in the 64x64 => 128 multiply in
mul_u64_add_u64_div_u64().
Link: https://lkml.kernel.org/r/20251105201035.64043-8-david.laight.linux@gmail.com
Signed-off-by: David Laight <david.laight.linux@gmail.com>
Reviewed-by: Nicolas Pitre <npitre@baylibre.com>
Cc: Biju Das <biju.das.jz@bp.renesas.com>
Cc: Borislav Betkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleinxer <tglx@linutronix.de>
Cc: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
If the product is only 64bits div64_u64() can be used for the divide.
Replace the pre-multiply check (ilog2(a) + ilog2(b) <= 62) with a simple
post-multiply check that the high 64bits are zero.
This has the advantage of being simpler, more accurate and less code. It
will always be faster when the product is larger than 64bits.
Most 64bit cpu have a native 64x64=128 bit multiply, this is needed (for
the low 64bits) even when div64_u64() is called - so the early check gains
nothing and is just extra code.
32bit cpu will need a compare (etc) to generate the 64bit ilog2() from two
32bit bit scans - so that is non-trivial. (Never mind the mess of x86's
'bsr' and any oddball cpu without fast bit-scan instructions.) Whereas the
additional instructions for the 128bit multiply result are pretty much one
multiply and two adds (typically the 'adc $0,%reg' can be run in parallel
with the instruction that follows).
The only outliers are 64bit systems without 128bit mutiply and simple in
order 32bit ones with fast bit scan but needing extra instructions to get
the high bits of the multiply result. I doubt it makes much difference to
either, the latter is definitely not mainstream.
If anyone is worried about the analysis they can look at the generated
code for x86 (especially when cmov isn't used).
Link: https://lkml.kernel.org/r/20251105201035.64043-4-david.laight.linux@gmail.com
Signed-off-by: David Laight <david.laight.linux@gmail.com>
Reviewed-by: Nicolas Pitre <npitre@baylibre.com>
Cc: Biju Das <biju.das.jz@bp.renesas.com>
Cc: Borislav Betkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleinxer <tglx@linutronix.de>
Cc: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "Implement mul_u64_u64_div_u64_roundup()", v5.
The pwm-stm32.c code wants a 'rounding up' version of
mul_u64_u64_div_u64(). This can be done simply by adding 'divisor - 1' to
the 128bit product. Implement mul_u64_add_u64_div_u64(a, b, c, d) = (a *
b + c)/d based on the existing code. Define mul_u64_u64_div_u64(a, b, d)
as mul_u64_add_u64_div_u64(a, b, 0, d) and mul_u64_u64_div_u64_roundup(a,
b, d) as mul_u64_add_u64_div_u64(a, b, d-1, d).
Only x86-64 has an optimsed (asm) version of the function. That is
optimised to avoid the 'add c' when c is known to be zero. In all other
cases the extra code will be noise compared to the software divide code.
The test module has been updated to test mul_u64_u64_div_u64_roundup() and
also enhanced it to verify the C division code on x86-64 and the 32bit
division code on 64bit.
This patch (of 9):
Change to prototype from mul_u64_u64_div_u64(u64 a, u64 b, u64 c) to
mul_u64_u64_div_u64(u64 a, u64 b, u64 d). Using 'd' for 'divisor' makes
more sense.
An upcoming change adds a 'c' parameter to calculate (a * b + c)/d.
Link: https://lkml.kernel.org/r/20251105201035.64043-1-david.laight.linux@gmail.com
Link: https://lkml.kernel.org/r/20251105201035.64043-2-david.laight.linux@gmail.com
Signed-off-by: David Laight <david.laight.linux@gmail.com>
Reviewed-by: Nicolas Pitre <npitre@baylibre.com>
Cc: Biju Das <biju.das.jz@bp.renesas.com>
Cc: Borislav Betkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleinxer <tglx@linutronix.de>
Cc: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "Optimize GCD performance on RISC-V by selecting
implementation at runtime", v3.
The current implementation of gcd() selects between the binary GCD and the
odd-even GCD algorithm at compile time, depending on whether
CONFIG_CPU_NO_EFFICIENT_FFS is set. On platforms like RISC-V, however,
this compile-time decision can be misleading: even when the compiler emits
ctz instructions based on the assumption that they are efficient (as is
the case when CONFIG_RISCV_ISA_ZBB is enabled), the actual hardware may
lack support for the Zbb extension. In such cases, ffs() falls back to a
software implementation at runtime, making the binary GCD algorithm
significantly slower than the odd-even variant.
To address this, we introduce a static key to allow runtime selection
between the binary and odd-even GCD implementations. On RISC-V, the
kernel now checks for Zbb support during boot. If Zbb is unavailable, the
static key is disabled so that gcd() consistently uses the more efficient
odd-even algorithm in that scenario. Additionally, to further reduce code
size, we select CONFIG_CPU_NO_EFFICIENT_FFS automatically when
CONFIG_RISCV_ISA_ZBB is not enabled, avoiding compilation of the unused
binary GCD implementation entirely on systems where it would never be
executed.
This series ensures that the most efficient GCD algorithm is used in
practice and avoids compiling unnecessary code based on hardware
capabilities and kernel configuration.
This patch (of 3):
On platforms like RISC-V, the compiler may generate hardware FFS
instructions even if the underlying CPU does not actually support them.
Currently, the GCD implementation is chosen at compile time based on
CONFIG_CPU_NO_EFFICIENT_FFS, which can result in suboptimal behavior on
such systems.
Introduce a static key, efficient_ffs_key, to enable runtime selection
between the binary GCD (using ffs) and the odd-even GCD implementation.
This allows the kernel to default to the faster binary GCD when FFS is
efficient, while retaining the ability to fall back when needed.
Link: https://lkml.kernel.org/r/20250606134758.1308400-1-visitorckw@gmail.com
Link: https://lkml.kernel.org/r/20250606134758.1308400-2-visitorckw@gmail.com
Co-developed-by: Yu-Chun Lin <eleanor15x@gmail.com>
Signed-off-by: Yu-Chun Lin <eleanor15x@gmail.com>
Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Ching-Chun (Jim) Huang <jserv@ccns.ncku.edu.tw>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Alexandre Ghiti <alexghiti@rivosinc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Be sure to test the extreme cases with and without bias.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Pull non-MM updates from Andrew Morton:
"Many singleton patches - please see the various changelogs for
details.
Quite a lot of nilfs2 work this time around.
Notable patch series in this pull request are:
- "mul_u64_u64_div_u64: new implementation" by Nicolas Pitre, with
assistance from Uwe Kleine-König. Reimplement mul_u64_u64_div_u64()
to provide (much) more accurate results. The current implementation
was causing Uwe some issues in the PWM drivers.
- "xz: Updates to license, filters, and compression options" from
Lasse Collin. Miscellaneous maintenance and kinor feature work to
the xz decompressor.
- "Fix some GDB command error and add some GDB commands" from
Kuan-Ying Lee. Fixes and enhancements to the gdb scripts.
- "treewide: add missing MODULE_DESCRIPTION() macros" from Jeff
Johnson. Adds lots of MODULE_DESCRIPTIONs, thus fixing lots of
warnings about this.
- "nilfs2: add support for some common ioctls" from Ryusuke Konishi.
Adds various commonly-available ioctls to nilfs2.
- "This series fixes a number of formatting issues in kernel doc
comments" from Ryusuke Konishi does that.
- "nilfs2: prevent unexpected ENOENT propagation" from Ryusuke
Konishi. Fix issues where -ENOENT was being unintentionally and
inappropriately returned to userspace.
- "nilfs2: assorted cleanups" from Huang Xiaojia.
- "nilfs2: fix potential issues with empty b-tree nodes" from Ryusuke
Konishi fixes some issues which can occur on corrupted nilfs2
filesystems.
- "scripts/decode_stacktrace.sh: improve error reporting and
usability" from Luca Ceresoli does those things"
* tag 'mm-nonmm-stable-2024-09-21-07-52' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (103 commits)
list: test: increase coverage of list_test_list_replace*()
list: test: fix tests for list_cut_position()
proc: use __auto_type more
treewide: correct the typo 'retun'
ocfs2: cleanup return value and mlog in ocfs2_global_read_info()
nilfs2: remove duplicate 'unlikely()' usage
nilfs2: fix potential oob read in nilfs_btree_check_delete()
nilfs2: determine empty node blocks as corrupted
nilfs2: fix potential null-ptr-deref in nilfs_btree_insert()
user_namespace: use kmemdup_array() instead of kmemdup() for multiple allocation
tools/mm: rm thp_swap_allocator_test when make clean
squashfs: fix percpu address space issues in decompressor_multi_percpu.c
lib: glob.c: added null check for character class
nilfs2: refactor nilfs_segctor_thread()
nilfs2: use kthread_create and kthread_stop for the log writer thread
nilfs2: remove sc_timer_task
nilfs2: do not repair reserved inode bitmap in nilfs_new_inode()
nilfs2: eliminate the shared counter and spinlock for i_generation
nilfs2: separate inode type information from i_state field
nilfs2: use the BITS_PER_LONG macro
...
Adds test suite for integer based power function which performs integer
exponentiation.
The test suite is designed to verify that the implementation of int_pow
correctly computes the power of a given base raised to a given exponent.
The tests check various scenarios and edge cases to ensure the accuracy
and reliability of the exponentiation function.
Updated commit with test information at commit time: Shuah Khan
Signed-off-by: Luis Felipe Hernandez <luis.hernandez093@gmail.com>
Reviewed-by: David Gow <davidgow@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Patch series "mul_u64_u64_div_u64: new implementation", v3.
This provides an implementation for mul_u64_u64_div_u64() that always
produces exact results.
This patch (of 2):
Library facilities must always return exact results. If the caller may be
contented with approximations then it should do the approximation on its
own.
In this particular case the comment in the code says "the algorithm
... below might lose some precision". Well, if you try it with e.g.:
a = 18446462598732840960
b = 18446462598732840960
c = 18446462598732840961
then the produced answer is 0 whereas the exact answer should be
18446462598732840959. This is _some_ precision lost indeed!
Let's reimplement this function so it always produces the exact result
regardless of its inputs while preserving existing fast paths when
possible.
Uwe said:
: My personal interest is to get the calculations in pwm drivers right.
: This function is used in several drivers below drivers/pwm/ . With the
: errors in mul_u64_u64_div_u64(), pwm consumers might not get the
: settings they request. Although I have to admit that I'm not aware it
: breaks real use cases (because typically the periods used are too short
: to make the involved multiplications overflow), but I pretty sure am
: not aware of all usages and it breaks testing.
:
: Another justification is commits like
: https://git.kernel.org/tip/77baa5bafcbe1b2a15ef9c37232c21279c95481c,
: where people start to work around the precision shortcomings of
: mul_u64_u64_div_u64().
Link: https://lkml.kernel.org/r/20240707190648.1982714-1-nico@fluxnic.net
Link: https://lkml.kernel.org/r/20240707190648.1982714-2-nico@fluxnic.net
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Tested-by: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Reviewed-by: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Tested-by: Biju Das <biju.das.jz@bp.renesas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pull non-MM updates from Andrew Morton:
- In the series "treewide: Refactor heap related implementation",
Kuan-Wei Chiu has significantly reworked the min_heap library code
and has taught bcachefs to use the new more generic implementation.
- Yury Norov's series "Cleanup cpumask.h inclusion in core headers"
reworks the cpumask and nodemask headers to make things generally
more rational.
- Kuan-Wei Chiu has sent along some maintenance work against our
sorting library code in the series "lib/sort: Optimizations and
cleanups".
- More library maintainance work from Christophe Jaillet in the series
"Remove usage of the deprecated ida_simple_xx() API".
- Ryusuke Konishi continues with the nilfs2 fixes and clanups in the
series "nilfs2: eliminate the call to inode_attach_wb()".
- Kuan-Ying Lee has some fixes to the gdb scripts in the series "Fix
GDB command error".
- Plus the usual shower of singleton patches all over the place. Please
see the relevant changelogs for details.
* tag 'mm-nonmm-stable-2024-07-21-15-07' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (98 commits)
ia64: scrub ia64 from poison.h
watchdog/perf: properly initialize the turbo mode timestamp and rearm counter
tsacct: replace strncpy() with strscpy()
lib/bch.c: use swap() to improve code
test_bpf: convert comma to semicolon
init/modpost: conditionally check section mismatch to __meminit*
init: remove unused __MEMINIT* macros
nilfs2: Constify struct kobj_type
nilfs2: avoid undefined behavior in nilfs_cnt32_ge macro
math: rational: add missing MODULE_DESCRIPTION() macro
lib/zlib: add missing MODULE_DESCRIPTION() macro
fs: ufs: add MODULE_DESCRIPTION()
lib/rbtree.c: fix the example typo
ocfs2: add bounds checking to ocfs2_check_dir_entry()
fs: add kernel-doc comments to ocfs2_prepare_orphan_dir()
coredump: simplify zap_process()
selftests/fpu: add missing MODULE_DESCRIPTION() macro
compiler.h: simplify data_race() macro
build-id: require program headers to be right after ELF header
resource: add missing MODULE_DESCRIPTION()
...
The number of times yet another open coded
`BITS_TO_LONGS(nbits) * sizeof(long)` can be spotted is huge.
Some generic helper is long overdue.
Add one, bitmap_size(), but with one detail.
BITS_TO_LONGS() uses DIV_ROUND_UP(). The latter works well when both
divident and divisor are compile-time constants or when the divisor
is not a pow-of-2. When it is however, the compilers sometimes tend
to generate suboptimal code (GCC 13):
48 83 c0 3f add $0x3f,%rax
48 c1 e8 06 shr $0x6,%rax
48 8d 14 c5 00 00 00 00 lea 0x0(,%rax,8),%rdx
%BITS_PER_LONG is always a pow-2 (either 32 or 64), but GCC still does
full division of `nbits + 63` by it and then multiplication by 8.
Instead of BITS_TO_LONGS(), use ALIGN() and then divide by 8. GCC:
8d 50 3f lea 0x3f(%rax),%edx
c1 ea 03 shr $0x3,%edx
81 e2 f8 ff ff 1f and $0x1ffffff8,%edx
Now it shifts `nbits + 63` by 3 positions (IOW performs fast division
by 8) and then masks bits[2:0]. bloat-o-meter:
add/remove: 0/0 grow/shrink: 20/133 up/down: 156/-773 (-617)
Clang does it better and generates the same code before/after starting
from -O1, except that with the ALIGN() approach it uses %edx and thus
still saves some bytes:
add/remove: 0/0 grow/shrink: 9/133 up/down: 18/-538 (-520)
Note that we can't expand DIV_ROUND_UP() by adding a check and using
this approach there, as it's used in array declarations where
expressions are not allowed.
Add this helper to tools/ as well.
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Acked-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David Laight
Kuan-Wei Chiu
Nicolas Pitre
Tamir Duberstein
Yu-Chun Lin
Bruno Sobreira França
Luis Felipe Hernandez
Linus Torvalds
Jeff Johnson
Alexander Lobakin