exponent bits of the reduced result, construct 2**k (hopefully in
parallel with the construction of the reduced result) and multiply by
it. This tends to be much faster if the construction of 2**k is
actually in parallel, and might be faster even with no parallelism
since adjustment of the exponent requires a read-modify-wrtite at an
unfortunate time for pipelines.
In some cases involving exp2* on amd64 (A64), this change saves about
40 cycles or 30%. I think it is inherently only about 12 cycles faster
in these cases and the rest of the speedup is from partly-accidentally
avoiding compiler pessimizations (the construction of 2**k is now
manually scheduled for good results, and -O2 doesn't always mess this
up). In most cases on amd64 (A64) and i386 (A64) the speedup is about
20 cycles. The worst case that I found is expf on ia64 where this
change is a pessimization of about 10 cycles or 5%. The manual
scheduling for plain exp[f] is harder and not as tuned.
Details specific to expm1*:
- the saving is closer to 12 cycles than to 40 for expm1* on i386 (A64).
For some reason it is much larger for negative args.
- also convert to __FBSDID().
the provided lock or &blocked_lock. The thread may be temporarily
assigned to the blocked_lock by the scheduler so a direct comparison
can not always be made.
- Use THREAD_LOCKPTR_ASSERT() in the primary consumers of the scheduling
interfaces. The schedulers themselves still use more explicit asserts.
Sponsored by: Nokia
exponent bits of the reduced result, construct 2**k (hopefully in
parallel with the construction of the reduced result) and multiply by
it. This tends to be much faster if the construction of 2**k is
actually in parallel, and might be faster even with no parallelism
since adjustment of the exponent requires a read-modify-wrtite at an
unfortunate time for pipelines.
In some cases involving exp2* on amd64 (A64), this change saves about
40 cycles or 30%. I think it is inherently only about 12 cycles faster
in these cases and the rest of the speedup is from partly-accidentally
avoiding compiler pessimizations (the construction of 2**k is now
manually scheduled for good results, and -O2 doesn't always mess this
up). In most cases on amd64 (A64) and i386 (A64) the speedup is about
20 cycles. The worst case that I found is expf on ia64 where this
change is a pessimization of about 10 cycles or 5%. The manual
scheduling for plain exp[f] is harder and not as tuned.
This change ld128/s_exp2l.c has not been tested.
obtained from OpenBSD with an algorithm suggested
by Amit Klein. The OpenBSD algorithm has a few
flaws; see Amit's paper for more information.
For a description of how this algorithm works,
please see the comments within the code.
Note that this commit does not yet enable random IP ID
generation by default. There are still some concerns
that doing so will adversely affect performance.
Reviewed by: rwatson
MFC After: 2 weeks
that is specialized for float precision. The new polynomial has degree
5 instead of 11, and a maximum error of 2**-27.74 ulps instead
of 2**-30.64. This doesn't affect the final error significantly; the
maximum error was and is about 0.9101 ulps on amd64 -01 and the number
of cases with an error of > 0.5 ulps is actually reduced by epsilon
despite the larger error in the polynomial.
This is about 15% faster on amd64 (A64), i386 (A64) and ia64. The asm
version is still used instead of this on i386 since it is faster and
more accurate.
threshold, according to the 'F' MALLOC_OPTIONS flag. This obsoletes the
'H' flag.
Try to realloc() large objects in place. This substantially speeds up
incremental large reallocations in the common case.
Fix a bug in arena_ralloc() that caused relocation of sub-page objects
even if the old and new sizes were in the same size class.
Maintain trees of runs and simplify the per-chunk page map. This allows
logarithmic-time searching for sufficiently large runs in
arena_run_alloc(), whereas the previous algorithm required linear time
in the worst case.
Break various large functions into smaller sub-functions, and inline
only the functions that are in the fast path for small object
allocation/deallocation.
Remove an unnecessary check in base_pages_alloc_mmap().
Avoid integer division in choose_arena() for the NO_TLS case on
single-CPU systems.
- Move recursion checking into rwlock inlines to free a bit for use with
adaptive spinners.
- Clear the RW_LOCK_WRITE_SPINNERS flag whenever the lock state changes
causing write spinners to restart their loop.
- Write spinners are limited by a count while readers hold the lock as
there is no way to know for certain whether readers are running still.
- In the read path block if there are write waiters or spinners to avoid
starving writers. Use a new per-thread count, td_rw_rlocks, to skip
starvation avoidance if it might cause a deadlock.
- Remove or change invalid assertions in turnstiles.
Reviewed by: attilio (developed parts of the patch as well)
Sponsored by: Nokia
This support tries to be as parallel as possible with other locking
primitives, but there are differences; more specifically:
- The base witness support is alredy equipped for allowing lock
duplication acquisition as lockmgr rely on this.
- In the case of lockmgr_disown() the lock result unlocked by witness
even if it is still held by the "kernel context"
- In the case of upgrading we can have 3 different situations:
* Total unlocking of the shared lock and nothing else
* Real witness upgrade if the owner is the first upgrader
* Shared unlocking and exclusive locking if the owner is not the first
upgrade but it is still allowed to upgrade
- LK_DRAIN is basically handled like an exclusive acquisition
Additively new options LK_NODUP and LK_NOWITNESS can now be used with
lockinit(): LK_NOWITNESS disables WITNESS for the specified lock while
LK_NODUP enable duplicated locks tracking. This will require manpages
update and a __FreeBSD_version bumping (addressed by further commits).
This patch also fixes a problem occurring if a lockmgr is held in
exclusive mode and the same owner try to acquire it in shared mode:
currently there is a spourious shared locking acquisition while what
we really want is a lock downgrade. Probabilly, this situation can be
better served with a EDEADLK failing errno return.
Side note: first testing on this patch alredy reveleated several LORs
reported, so please expect LORs cascades until resolved. NTFS also is
reported broken by WITNESS introduction. BTW, NTFS is exposing a lock
leak which needs to be fixed, and this patch can help it out if
rightly tweaked.
Tested by: kris, yar, Scot Hetzel <swhetzel at gmail dot com>
done in consumers code: using locks properties is much more appropriate.
Fix current code doing these bogus checks.
Note: Really, callout are not usable by all !(LC_SPINLOCK | LC_SLEEPABLE)
primitives like rmlocks doesn't implement the generic lock layer
functions, but they can be equipped for this, so the check is still
valid.
Tested by: matteo, kris (earlier version)
Reviewed by: jhb
