While going over various problems signaled by valgrind when running
`make check` on a build configured with `--enable-valgrind`, I
discovered a bug in this test case.
Namely, the test case was previously creating an `i32` (originally an
`int` aka `isize` but then we changed the name and the fallback
rules), and then reading from a `*const isize`. Valgrind rightly
complains about this, since we are reading an 8 byte value on 64-bit
systems, but in principle only 4 bytes have been initialized.
(I wish this was the only valgrind unclean test, but unfortunately
there are a bunch more. This was just the easiest/first one that I
dissected.)
This commit removes the call to `panic!("Some tests failed")` at the end of all
tests run when running with libtest. The panic is replaced with
`std::process::exit` to have a nonzero error code, but this change both:
1. Makes the test runner no longer print out the extraneous panic message at the
end of a failing test run that some tests failed. (this is already summarized
in the output of the test run).
2. When running tests with `RUST_BACKTRACE` set it removes an extraneous
backtrace from the output (only failing tests will have their backtraces in
the output.
also, use the right caching logic for type_moves_by_default (this was
broken by @jroesch).
```
before:
593.10user 5.21system 7:51.41elapsed 126%CPU (0avgtext+0avgdata 1150016maxresident)k
after:
567.03user 4.00system 7:28.23elapsed 127%CPU (0avgtext+0avgdata 1133112maxresident)k
```
A nice 4.5% improvement. For reference, on the last run LLVM takes 429.267s, which is 75% - hopefully this can be reduced.
I think the regression since #27751 is because of the wf patch - need to investigate it through.
r? @nikomatsakis
There wasn't any particular reason the functions needed to be there
anyway, so just get rid of them, and adjust libstd to compensate.
With this change, libcore depends on exactly two floating-point functions:
fmod and fmodf. They are implicitly referenced because they are used to
implement "%".
Dependencies of libcore on Linux x86-x64 with this patch:
```
0000000000000000 *UND* 0000000000000000 __powidf2
0000000000000000 *UND* 0000000000000000 __powisf2
0000000000000000 *UND* 0000000000000000 fmod
0000000000000000 *UND* 0000000000000000 fmodf
0000000000000000 *UND* 0000000000000000 memcmp
0000000000000000 *UND* 0000000000000000 memcpy
0000000000000000 *UND* 0000000000000000 memset
0000000000000000 *UND* 0000000000000000 rust_begin_unwind
0000000000000000 *UND* 0000000000000000 rust_eh_personality
```
StrSearcher: Implement the complete reverse case for the two way algorithm
Fix quadratic behavior in StrSearcher in reverse search with periodic
needles.
This commit adds the missing pieces for the "short period" case in
reverse search. The short case will show up when the needle is literally
periodic, for example "abababab".
Two way uses a "critical factorization" of the needle: x = u v.
Searching matches v first, if mismatch at character k, skip k forward.
Matching u, if mismatch, skip period(x) forward.
To avoid O(mn) behavior after mismatch in u, memorize the already
matched prefix.
The short period case requires that |u| < period(x).
For the reverse search we need to compute a different critical
factorization x = u' v' where |v'| < period(x), because we are searching
for the reversed needle. A short v' also benefits the algorithm in
general.
The reverse critical factorization is computed quickly by using the same
maximal suffix algorithm, but terminating as soon as we have a location
with local period equal to period(x).
This adds extra fields crit_pos_back and memory_back for the reverse
case. The new overhead for TwoWaySearcher::new is low, and additionally
I think the "short period" case is uncommon in many applications of
string search.
The maximal_suffix methods were updated in documentation and the
algorithms updated to not use !0 and wrapping add, variable left is now
1 larger, offset 1 smaller.
Use periodicity when computing byteset: in the periodic case, just
iterate over one period instead of the whole needle.
Example before (rfind) after (twoway_rfind) benchmark shows the removal
of quadratic behavior.
needle: "ab" * 100, haystack: ("bb" + "ab" * 100) * 100
```
test periodic::rfind ... bench: 1,926,595 ns/iter (+/- 11,390) = 10 MB/s
test periodic::twoway_rfind ... bench: 51,740 ns/iter (+/- 66) = 386 MB/s
```
This implements https://github.com/rust-lang/rfcs/pull/1199 (except for doing all the platform intrinsics).
Things remaining for SIMD (not necessarily in this PR):
- [x] I (@huonw) am signed up to ensure the compiler matches the RFC, when it lands
- [x] the platform specific intrinsics aren't properly type checked at the moment (LLVM will throw a "random" assertion)
- [ ] there's a lot of useful intrinsics that are missing, including whole platforms (mips, powerpc)
- [ ] the target-feature `cfg` detection/adding is not so great at the moment
- [x] I think the platform specific intrinsics should go in their own `extern` ABI (i.e. not `"rust-intrinsic"`)
(I'm adjusting the RFC to reflect the latter.)
I think it would be very nice for this to land without requiring the RFC to land first, because of the first point, and because this is the only way for any further work to happen/be experimented with, without requiring people to build/install/multirust a compiler from a custom branch.
r? @alexcrichton
This is purposely separate to the "rust-intrinsic" ABI, because these
intrinsics are theoretically going to become stable, and should be fine
to be independent of the compiler/language internals since they're
intimately to the platform.
This is theoretically a breaking change, but GitHub search turns up no
uses of it, and most non-built-in cfg's are passed via cargo features,
which look like `feature = "..."`, and hence can't overlap.
This commit removes the call to `panic!("Some tests failed")` at the end of all
tests run when running with libtest. The panic is replaced with
`std::process::exit` to have a nonzero error code, but this change both:
1. Makes the test runner no longer print out the extraneous panic message at the
end of a failing test run that some tests failed. (this is already summarized
in the output of the test run).
2. When running tests with `RUST_BACKTRACE` set it removes an extraneous
backtrace from the output (only failing tests will have their backtraces in
the output.
