I couldn't find a good way to spread `ip0` and `ip1` apart when we accelerate
due to incompressible inputs. (The methods I tried slowed things down quite a
bit.)
Since we aren't splaying ip0 and ip1 apart (which would be like `0_1_2_3_`, as
opposed to the `01__23__` we were actually doing), it's a big ambitious to
increment `step` by 2. Instead, let's increment it by 1, which has the benefit
sliiightly improving compression. Speed remains pretty much unchanged.
The position updates are rewritten from `ip[N] = ip[N-1] + step` to be
`ip[N] = ip[N-2] + step`. This lets us only deal with the asymmetric spacing
of gaps at setup and then we only have to keep a single `step` variable.
This seems to work quite well on GCC and Clang!
This replicates the behavior of @terrelln's `ZSTD_fast` implementation. That
is, it always looks at adjacent pairs of positions, and only applies the
acceleration every other position. This produces a more fine-grained
acceleration.
Use the same trick as we did for zstd_lazy in PR #2828:
* Create one search function specialization for each (dictMode, mls).
* Select the search function pointer at the top of the match finder.
Additionally, we no longer inline `ZSTD_compressBlock_opt_generic` into
every function, since `dictMode` is no longer used as a template. Create
two specializations, for opt levels 0 and 2, and call one of the two
specializations.
Lastly, remove the hack that disabled inlining for zstd_opt for the
Linux Kernel, as we've gotten most of the benefit already.
Compilation time sees a ~4x reduction:
| Compiler | Flags | Dev Time (s) | PR Time (s) | Delta |
|----------|----------------------------------|--------------|-------------|-------|
| gcc | -O3 | 10.1 | 2.3 | -77% |
| gcc | -O3 -fsanitize=address,undefined | 61.1 | 10.2 | -83% |
| clang | -O3 | 9.0 | 2.1 | -76% |
| clang | -O3 -fsanitize=address,undefined | 33.5 | 5.1 | -84% |
Build size is reduced by 150KB - 200KB:
| Compiler | Dev libzstd.a Size (B) | PR libzstd.a Size (B) | Delta |
|----------|------------------------|-----------------------|-------|
| gcc | 1327476 | 1177108 | -11% |
| clang | 1378324 | 1167780 | -15% |
There is a <2% speed loss in all cases:
| Compiler | Level | Dev Speed (MB/s) | PR Speed (MB/s) | Delta |
|----------|-------|------------------|-----------------|--------|
| gcc | 16 | 4.78 | 4.72 | -1.25% |
| gcc | 17 | 3.49 | 3.46 | -0.85% |
| gcc | 18 | 2.92 | 2.86 | -2.04% |
| gcc | 19 | 2.61 | 2.61 | 0.00% |
| clang | 16 | 4.69 | 4.80 | 2.34% |
| clang | 17 | 3.53 | 3.49 | -1.13% |
| clang | 18 | 2.86 | 2.85 | -0.34% |
| clang | 19 | 2.61 | 2.61 | 0.00% |
Fixes Issue #2862.
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`.
* When dynamic dispatching to bmi2 add lzcnt and bmi to the
TARGET_ATTRIBUTE.
* Centralize the bmi2 TARGET_ATTRIBUTE definition to
BMI2_TARGET_ATTRIBUTE so we can change it in the future.
* Only enable bmi2 when both bmi1 & bmi2 are supported. There shouldn't
be any cases where bmi2 is supported but bmi1 isn't. But, since we are
using the instruction we should check bmi1 as well.
PR #2850 attempted to fix a determinism bug that was uncovered by OSS-Fuzz. It
succeeded in addressing that source of non-determinism, but introduced a new
one: it was possible, when index reduction occurred, to map indices in the
window to the reserved value, which would cause them to be zeroed, potentially
altering parsing of the input.
This PR addresses this issue. It makes sure that the bottom of the window is
always `>= ZSTD_WINDOW_START_INDEX`.
I'm not sure if this makes #2850 redundant. I think it's probably still
valuable to have that protection as well.
Credit to OSS-Fuzz for discovering this issue.
The optimal parser is unlikely to be used in the linux kernel in
practice. There is no reason these functions should be force inlined,
since we aren't gaining anything, and are losing build size.
| Compiler | Before (Bytes) | After (Bytes) | Delta (Bytes) |
|----------|----------------|---------------|---------------|
| gcc-11 | 1142090 | 952754 | -189336 |
| clang-12 | 1228402 | 976290 | -252112 |
This is a temporary solution pending the resolution of PR #2862 in the
`dev` branch.
Take the same approach as in PR #2828 [0] to remove functions that force
inline many function bodies and `switch`. Instead, create one function per
"template" combination, and then switch between these functions. This
allows the compiler to break the large function into many small
functions, which generally helps codegen.
Also, in the `extDict` modes when there is no ext-dict, call the top
level function instead of the force inlined one, to save on code size.
I'm specifically doing this because gcc on the parisc architecture doesn't
handle the large function body well, and ends up using a lot of excess
stack space. Outlining these functions fixes it.
Previously, if an index was equal to `reducerValue + 1`, it would get remapped
during index reduction to 1 i.e. `ZSTD_DUBT_UNSORTED_MARK`. This can affect the
parsing of the input slightly, by causing tree nodes to be nullified when they
otherwise wouldn't be. This hardly matters from a correctness or efficiency
perspective, but it does impact determinism.
So this commit changes index reduction to avoid mapping indices to collide with
`ZSTD_DUBT_UNSORTED_MARK`.
that's clearer than finding the tables somewhere in the middle of `compress.c`.
Also, down the line, it may potentially allows zstd to feature adjusted tables depending on target cpu.
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/
There is no minimum value check, so the parameter could be negative.
Switch to the standard pattern of using `BOUNDCHECK()`.
The bug was reported by Dan Carpenter and found by Smatch static
checker.
https://lore.kernel.org/all/20211008063704.GA5370@kili/
