Age | Commit message (Collapse) | Author | Files | Lines |
|
Looks like CPU doesn't like overlaping loads.
This new codes in some cases shows 3-4 clock improvements.
Change-Id: Ia1b49976ad95140c573f892fdc0a32eebbfa06c8
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
If non-trivial alignment (e.g. 64) requested, and the object size
(e.g. 16) is smaller than (alignment_request -
MHEAP_ELT_OVERHEAD_BYTES), round up the size request.
This avoids creating remainder chunks, which are
false-cache-line-sharing bait to begin with.
Change-Id: Ie1a21286d29557d125bb346254b1be2def868b1a
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Change-Id: Ida678e6f31daa8decb18189da712a350336326e2
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Stop spending cycles repeatedly tail-trimming the pool free element
bitmap; possibly at the expense of slightly hurting pool_foreach
peformance.
Change-Id: I8a7f3e7b26c71d7496ba9393b2a167dc7f538355
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Change-Id: I4e0fd487970796f0153a5b16333827d23b57deac
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Add aarch64 neon intrinsics to fix build failures similar to this:
error: implicit declaration of function ‘u64x2_load_unaligned’
Change-Id: I6178504a48242742df3f7d75abdaf108796cf73f
Signed-off-by: Sirshak Das <sirshak.das@arm.com>
|
|
Change-Id: I16395bbf843e338cdd366d85bb4df3de95d9b265
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
crash stack backtrace will be directed to syslog
1. make use of glic backtrace in execinfo.h. the old clib_backtrace is removed
2. install SIGABRT in signal handler, but have to remove it when backtrace is
done. reason is to capture stack trace caused by SIGABRT. vPP ASSERT always
call os_exit then abort(). we definitely want to know the trace of this
situation. It is a little tricky to avoid SIGABRT infinite loop
3. always load symbols by calling clib_elf_main_init () in main(). Otherwise,
PC addresses instead of symbols will be displayed.
Change-Id: I150e15b94a4620b2ea4f08c73dc3e6ad1856de1e
Signed-off-by: Kingwel Xie <kingwel.xie@ericsson.com>
|
|
Change-Id: Ib3a55598a83cc99485b40e38e7c406ecb126fd42
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Because it avoids pool putting/getting the timer, this function is
somewhat faster than stopping and restarting a timer.
Change-Id: Id99ed9d356b0b1f7e12facfe8da193e1cd30b3ec
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Use similar approach as in the clib_bihash_search_inline_with_hash to
be able to do the hash calculation and lookup separately.
Change-Id: Ief79aa0f9f1e42b0af88be4807ca01fac30a80d7
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
It'll be interesting to see what the perf trend job
says about this change.
Change-Id: I66307a19a865011ac9660108098874fa1481c895
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Change-Id: I2356b1e05fd868b46b4d26ade760900a5739ca4d
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
if you plan to put a hash into shared memory, the key sum and key
equal functions MUST be set to constants such as KEY_FUNC_STRING,
KEY_FUNC_MEM, etc. -lvppinfra is PIC, which means that the process
which set up the hash won't have the same idea where the key sum and
key compare functions live in other processes.
Change-Id: Ib3b5963a0d2fb467b91e1f16274df66ac74009e9
Signed-off-by: Ole Troan <ot@cisco.com>
Signed-off-by: Dave Barach <dave@barachs.net>
Signed-off-by: Ole Troan <ot@cisco.com>
|
|
Added configure argument "--with-log2-cache-line-bytes=5|6|7|auto"
AKA 32, 64, or 128 bytes, or use the inferred value from the build host.
produces build-xxx/vpp/vppinfra/config.h, which .../src/vppinfra/cache.h
Kernels which implement the following pseudo-file (aka x86_64) are
easy: /sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size
Otherwise, extract the cpuid from /proc/cpuinfo and map it to the
cache line size.
Change-Id: I7ff861e042faf82c3901fa1db98864fbdea95b74
Signed-off-by: Dave Barach <dave@barachs.net>
Signed-off-by: Nitin Saxena <nitin.saxena@cavium.com>
|
|
Change-Id: Ib121b24935d5c706cfba6e4b6d321086a38cad91
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Fixes clang error: value size does not match register size specified
by the constraint and modifier
Change-Id: I83e69445eacd6570607334e086a8582addb5bdfc
Signed-off-by: Sirshak Das <sirshak.das@arm.com>
Reviewed-by: Brian Brooks <brian.brooks@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
|
|
This fixes some compilation warnings with clang on AArch64.
Change-Id: Idb941944e3f199f483c80e143a9e5163a031c4aa
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Ibab5e27277f618ceb2d543b9d6a1a5f191e7d1db
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Ia1b49d7fd5f32d9a5139df5df636b46264003a63
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Ieae4ff6429fc5bdcf0e243db40ab7ec00c30730a
Signed-off-by: Pierre Pfister <ppfister@cisco.com>
|
|
- hash is great. But it is a bit too slow for the DP. Use direct array indexing
to quickly retrieve the slave interface.
- the algorithm used by flow hash is great. But it is a bit too slow for the DP.
Use l2_hash_hash() extracted from lb_hash.h which ECMP is using. It makes use
of intrinsic crc32 instruction set.
- shortcut modulo arithmetic when the operand is 2**x (where x up to 4) to
avoid division instruction.
- special case for link count == 1 in bond_tx_fn()
- use clib_mem_unaligned to access data for the packet to avoid alignment error
- Fix some typos for packet tracing.
Change-Id: I8eae3ad497061c5473aa675ba894ee0211120d25
Signed-off-by: Steven <sluong@cisco.com>
|
|
bihash_48_8 case:
Scalar code: 6 clocks
SSE4.2 code: 3 clocks
AVX2 code: 2.27 clocks
AVX512 code: 1.5 clocks
Change-Id: I40700175835a1e7321276e47eadbf9771d3c5a68
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: I56782652d8ef10304900cc293cfc0502689d800e
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Remove functions which have native C equivalent (i.e. _is_equal can be
replaced with ==, _add with +)
Add SSE4.2, AVX-512 implementations of splat, load_unaligned, store_unaligned,
is_all_zero, is_equal, is_all_equal
Change-Id: Ie80b0e482e7a76248ad79399c2576468532354cd
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
This fixes ARM64 build where we dont have defined u16x8_msb_mask(...)
Change-Id: I864f5134a0d951601810c800f587d173b3b7ef41
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: I1042c0fe179b57a00ce99c8d62cb1bdbe24d9184
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Ib3fcc3ceb7f315389bcdecbb7d9632540a5dd6ba
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Scan IPv4 and IPv6 neigbor pool entries once a minute to keep them
up to date. The neighbor of an entry is probed if its time-stamp
is older than 1 minute. If the neighbor respond, its time-stamp
will be updated. If there is no response from a neighbor, its
entry will be deleted when the time-stamp of the entry become more
than 4 minutes old. Static neighbor entries are not probed nor
deleted.
Implemented CLI and API to enable and disable priodic scan of IPv4,
IPv6 or both types of IP neighbors. CLI is "ip scan-neighbor" and
API is "ip_scan_neighbor_enable_disable". Other IP neighbor scan
parameters can also be changed from their defaults via the CLI/API.
Change-Id: Id1a0a934ace15d03db845aa698bcbb9cdabebfcd
Signed-off-by: John Lo <loj@cisco.com>
|
|
Change-Id: I1f7c634328f25b33580a215af2daeb498cd3b181
Signed-off-by: Klement Sekera <ksekera@cisco.com>
|
|
It is much cheaper to use ctzll than to do shift,subtract and mask
in likely case when we are looking for 1st set bit in the uword.
Change-Id: I31954081571978878c7098bafad0c85a91755fa2
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Id02d613b8613a2d448840fe2d6a5e3b168a3c563
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Ifea9c772e8784642433b92091f5769eb9ec06890
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: I2896dbde78b5d58dc706756f4c76632c303557ae
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Object sizes must evenly divide alignment requests, or vice
versa. Otherwise, only the first object will be aligned as
requested.
Three choices: add CLIB_CACHE_LINE_ALIGN_MARK(align_me) at
the end of structures, manually pad to an even divisor or multiple of
the alignment request, or use plain vectors/pools.
static assert for enforcement.
Change-Id: I41aa6ff1a58267301d32aaf4b9cd24678ac1c147
Signed-off-by: Dave Barach <dbarach@cisco.com>
|
|
Change-Id: I54a00686a7f3a61f583a5f701a0ab6c5480a455b
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Change-Id: If174d189de40e6f9ffae99997bba93a2519d9fda
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: I6c1c855cf5fc2ee06f1c7ddd6576ca16cd556fdd
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: Icfb99a09726c01e96ff14967afbafa4116e02eff
Signed-off-by: Dave Barach <dbarach@cisco.com>
|
|
Change-Id: Ide8bf41e24a427643a3a17b1c9089993790c12a6
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Removed the sole use of it from ip6_neighbor.c
Change-Id: Ie53cb3b6a3a41ec0917ec2042e5006d0cfaefc01
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Use (u64) cast to ensure proper address calculations.
Change-Id: I6bad50010b140189f1b0af177e55da0045bd7a93
Signed-off-by: Lee Roberts <lee.roberts@hpe.com>
|
|
With glibc 2.27 the memfd_create has been added to the devel libraries.
That's causing the internally defined static function to clash with the
system wide one. This patch addresses that issue on systems with latest
glibc libraries.
Change-Id: I788bf49b23d5b5f1cb1c0374e243d8a429178a71
Signed-off-by: Marco Varlese <marco.varlese@suse.com>
|
|
Change-Id: I6d049c0875b91f67f008dc04ae7efe2f8ddc276e
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
platform. VPP-1129
Change-Id: I954acb56d901e42976e71534317f38d7c4359bcf
Signed-off-by: Adrian Oanca <adrian.oanca@enea.com>
|
|
Nintin
Change-Id: I88aabd34ef385d620695ac17ec3fe2f4a5177ada
Signed-off-by: Adrian Oanca <adrian.oanca@enea.com>
|
|
Change-Id: I2e9d01ccba5288e89b886464436097d3cb7d2d18
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
Directly allocate and carve cache-line-aligned chunks of virtual
memory. To a first approximation, bihash wasn't using
clib_mem_free(...).
We eliminate mheap object header/trailers, which improves space
efficiency. We also eliminate the 4gb bihash table size limit. An 8_8
bihash w/ 100 million random entries uses 3.8 Gbytes.
Change-Id: Icf925fdf99bce7d6ac407ac4edd30560b8f04808
Signed-off-by: Dave Barach <dave@barachs.net>
|
|
This change only affects Aarch64 where previously we were using 128
bytes.
Change-Id: I52a3f2f3ff8c06abe8ae3933bc0d7a2a7749dd8a
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: I71fac34b071a07b4331cb5c900c3b8667c1fc114
Signed-off-by: Damjan Marion <damarion@cisco.com>
|