Every 256 bytes the lazy match finders process without finding a match,
they will increase their step size by 1. So for bytes [0, 256) they search
every position, for bytes [256, 512) they search every other position,
and so on. However, they currently still insert every position into
their hash tables. This is different from fast & dfast, which only
insert the positions they search.
This PR changes that, so now after we've searched 2KB without finding
any matches, at which point we'll only be searching one in 9 positions,
we'll stop inserting every position, and only insert the positions we
search. The exact cutoff of 2KB isn't terribly important, I've just
selected a cutoff that is reasonably large, to minimize the impact on
"normal" data.
This PR only adds skipping to greedy, lazy, and lazy2, but does not
touch btlazy2.
| Dataset | Level | Compiler | CSize ∆ | Speed ∆ |
|---------|-------|--------------|---------|---------|
| Random | 5 | clang-14.0.6 | 0.0% | +704% |
| Random | 5 | gcc-12.2.0 | 0.0% | +670% |
| Random | 7 | clang-14.0.6 | 0.0% | +679% |
| Random | 7 | gcc-12.2.0 | 0.0% | +657% |
| Random | 12 | clang-14.0.6 | 0.0% | +1355% |
| Random | 12 | gcc-12.2.0 | 0.0% | +1331% |
| Silesia | 5 | clang-14.0.6 | +0.002% | +0.35% |
| Silesia | 5 | gcc-12.2.0 | +0.002% | +2.45% |
| Silesia | 7 | clang-14.0.6 | +0.001% | -1.40% |
| Silesia | 7 | gcc-12.2.0 | +0.007% | +0.13% |
| Silesia | 12 | clang-14.0.6 | +0.011% | +22.70% |
| Silesia | 12 | gcc-12.2.0 | +0.011% | -6.68% |
| Enwik8 | 5 | clang-14.0.6 | 0.0% | -1.02% |
| Enwik8 | 5 | gcc-12.2.0 | 0.0% | +0.34% |
| Enwik8 | 7 | clang-14.0.6 | 0.0% | -1.22% |
| Enwik8 | 7 | gcc-12.2.0 | 0.0% | -0.72% |
| Enwik8 | 12 | clang-14.0.6 | 0.0% | +26.19% |
| Enwik8 | 12 | gcc-12.2.0 | 0.0% | -5.70% |
The speed difference for clang at level 12 is real, but is probably
caused by some sort of alignment or codegen issues. clang is
significantly slower than gcc before this PR, but gets up to parity with
it.
I also measured the ratio difference for the HC match finder, and it
looks basically the same as the row-based match finder. The speedup on
random data looks similar. And performance is about neutral, without the
big difference at level 12 for either clang or gcc.
Part 2 of #3528
Adds hash salt that helps to avoid regressions where consecutive compressions use the same tag space with similar data (running zstd -b5e7 enwik8 -B128K reproduces this regression).
#3543 decreases the size of the tagTable by a factor of 2, which requires using the first tag position in each row for head position instead of a tag.
Although position 0 stopped being a valid match, it still persisted in mask calculation resulting in the matches loops possibly terminating before it should have. The fix skips position 0 to solve this problem.
Allocate half the memory for tag space, which means that we get one less slot for an actual tag (needs to be used for next position index).
The results is a slight loss in compression ratio (up to 0.2%) and some regressions/improvements to speed depending on level and sample. In turn, we get to save 16% of the hash table's space (5 bytes per entry instead of 6 bytes per entry).
* Cap shortCache chainLog to 24
* Cap row match finder hashLog so that rowLog <= 24
* Add unit tests to expose all cases. The row match finder unit tests
are only run in 64-bit mode, because they allocate ~1GB.
Fixes#3336
```
for f in $(find . \( -path ./.git -o -path ./tests/fuzz/corpora \) -prune -o -type f);
do
sed -i 's/Facebook, Inc\./Meta Platforms, Inc. and affiliates./' $f;
done
```
Use a switch statement to select the search function instead of an
indirect function call. This results in a sizable performance win.
This PR is a modification of the approach taken in PR #2828.
When I measured performance for that commit, it was neutral.
However, I now see a performance regression on gcc, but still
neutral on clang. I'm measuring on the same platform, but with
newer compilers. The new approach beats both the current dev
branch and the baseline before PR #2828 was merged.
This PR is necessary for Issue #3275, to update zstd in the kernel.
Without this PR there is a large regression in greedy - btlazy2
compression speed. With this PR it is about neutral.
gcc version: 12.2.0
clang version: 14.0.6
dataset: silesia.tar
| Compiler | Level | Dev Speed (MB/s) | PR Speed (MB/s) | Delta |
|----------|-------|------------------|-----------------|--------|
| gcc | 5 | 102.6 | 113.7 | +10.8% |
| gcc | 7 | 66.6 | 74.8 | +12.3% |
| gcc | 9 | 51.5 | 58.9 | +14.3% |
| gcc | 13 | 14.3 | 14.3 | +0.0% |
| clang | 5 | 108.1 | 114.8 | +6.2% |
| clang | 7 | 68.5 | 72.3 | +5.5% |
| clang | 9 | 53.2 | 56.2 | +5.6% |
| clang | 13 | 14.3 | 14.7 | +2.8% |
The binary size stays just about the same for clang and gcc, measured
using the `size` command:
| Compiler | Branch | Text | Data | BSS | Total |
|----------|--------|---------|------|-----|---------|
| gcc | dev | 1127950 | 3312 | 280 | 1131542 |
| gcc | PR | 1123422 | 2512 | 280 | 1126214 |
| clang | dev | 1046254 | 3256 | 216 | 1049726 |
| clang | PR | 1048198 | 2296 | 216 | 1050710 |
Comparing 4B instead of comparing 1B in ZSTD_noDict
mode, thus it can avoid cases like match in match[ml]
but mismatch in match[ml-3]..match[ml-1]. So the call
count of ZSTD_count can be reduced.
Signed-off-by: Jun He <jun.he@arm.com>
Change-Id: I3449ea423d5c8e8344f75341f19a2d1643c703f6
We found that movemask is not used properly or consumes too much CPU.
This effort helps to optimize the movemask emulation on ARM.
For level 8-9 we saw 3-5% improvements. For level 10 we say 1.5%
improvement.
The key idea is not to use pure movemasks but to have groups of bits.
For rowEntries == 16, 32 we are going to have groups of size 4 and 2
respectively. It means that each bit will be duplicated within the group
Then we do AND to have only one bit set in the group so that iteration
with lowering bit `a &= (a - 1)` works as well.
Also, aarch64 does not have rotate instructions for 16 bit, only for 32
and 64, that's why we see more improvements for level 8-9.
vshrn_n_u16 instruction is used to achieve that: vshrn_n_u16 shifts by
4 every u16 and narrows to 8 lower bits. See the picture below. It's
also used in
[Folly](c570259008/folly/container/detail/F14Table.h (L446)).
It also uses 2 cycles according to Neoverse-N{1,2} guidelines.
64 bit movemask is already well optimized. We have ongoing experiments
but were not able to validate other implementations work reliably faster.
this meant to abstract the sumtype representation required
to transfert `offcode` to `ZSTD_storeSeq()`.
Unfortunately, the sumtype numeric representation is currently a leaky abstraction
that has permeated many other parts of the code,
especially within `zstd_lazy.c` and also within `zstd_opt.c` and `zstd_compress.c`.
While this PR makes a good job a transfering a large nb of call sites
to using the new macros, there are still a few sites where this transformation is more complex,
or where the numeric representation itself it used "as is".
One of the problematics area is the decision to use the numeric format of the sumtype
within the match finders of `zstd_lazy`.
This commit doesn't change the behavior, it only introduces and employes the macros,
but eventually the resulting code remains identical.
At target, if the numeric representation of the sumtype can be completely abstracted
and no other part of the code depends on it,
it will be possible to move it towards something slightly more efficient.
the variable has only very limited usage,
being only used once at the beginning of the block for prefetching only,
hence the error had no impact on compression ratio.
mostly for maintenance convenience.
Performance wise, there is very little change,
slightly faster for slog 3 & 4,
neutral or very slightly negative for slot 5 & 6.
short-tests-0 were silently failing. I think because of the && make clean construction. Switch to ; instead.
Also fix all the test failures that were exposed.
`make all` is failing on CircleCI because it is missing Docker. Move that test
to GitHub actions, and switch the pedantic CircleCI test to `make allmost`.
Speed up compilation times by moving each specialized search function
into its own function. This is faster because compilers can handle many
smaller functions much faster than one gigantic function. The previous
approach generated one giant function with `switch` statements and
inlining to select the implementation.
| Compiler | Flags | Dev Time (s) | PR Time (s) | Delta |
|----------|-------------------------------------|--------------|-------------|-------|
| gcc | -O3 | 16.5 | 5.6 | -66% |
| gcc | -O3 -g -fsanitize=address,undefined | 158.9 | 38.2 | -75% |
| clang | -O3 | 36.5 | 5.5 | -85% |
| clang | -O3 -g -fsanitize=address,undefined | 27.8 | 17.5 | -37% |
This also reduces the binary size because the search functions are no
longer inlined into the main body.
| Compiler | Dev libzstd.a Size (B) | PR libzstd.a Size (B) | Delta |
|----------|------------------------|-----------------------|-------|
| gcc | 1563868 | 1308844 | -16% |
| clang | 1924372 | 1376020 | -28% |
Finally, the performance is not impacted significantly by this change,
in fact we generally see a small speed boost.
| Compiler | Level | Dev Speed (MB/s) | PR Speed (MB/s) | Delta |
|----------|-------|------------------|-----------------|-------|
| gcc | 5 | 110.6 | 110.0 | -0.5% |
| gcc | 7 | 70.4 | 72.2 | +2.5% |
| gcc | 9 | 53.2 | 53.5 | +0.5% |
| gcc | 13 | 12.7 | 12.9 | +1.5% |
| clang | 5 | 113.9 | 110.4 | -3.0% |
| clang | 7 | 67.7 | 70.6 | +4.2% |
| clang | 9 | 51.9 | 52.2 | +0.5% |
| clang | 13 | 12.4 | 13.3 | +7.2% |
The compression strategy is unmodified in this PR, so the compressed size
should be exactly the same. I may have a follow up PR to slightly improve
the compression ratio, if it doesn't cost too much speed.
Fix underflow of `nbCompares` by switching to an `int` and comparing
`nbCompares > 0`. This is a minimal fix, because I don't want to change
the logic. These loops seem to be doing `nbCompares + 1` comparisons.
The bug was reported by Dan Carpenter and found by Smatch static
checker.
https://lore.kernel.org/all/20211008063704.GA5370@kili/
turns out, it's possible to constify MatchState* parameter
in some parts of the binary tree algorithm,
making it a pure read-only parameter,
as opposed to a mutable state.
This is supposed to be helpful for both maintenance and the compiler.
