This improves the spectralnorm shootout benchmark through a few vectors after
looking at the leading C implementation:
* The simd-based f64x2 is now used to parallelize a few computations
* RWLock usage has been removed. A custom `parallel` function was added as a
form of stack-based fork-join parallelism. I found that the contention on the
locks was high as well as hindering other optimizations.
This does, however, introduce one `unsafe` block into the benchmarks, which
previously had none.
In terms of timings, the before and after numbers are:
```
$ time ./shootout-spectralnorm-before
./shootout-spectralnorm-before 2.07s user 0.71s system 324% cpu 0.857 total
$ time ./shootout-spectralnorm-before 5500
./shootout-spectralnorm-before 5500 11.88s user 1.13s system 459% cpu 2.830 total
$ time ./shootout-spectralnorm-after
./shootout-spectralnorm-after 0.58s user 0.01s system 280% cpu 0.210 tota
$ time ./shootout-spectralnorm-after 5500
./shootout-spectralnorm-after 5500 3.55s user 0.01s system 455% cpu 0.783 total
```
Doing so would incur deeply nested expansion of the tree with no useful
side effects. This is problematic for "wide" data types such as structs
with dozens of fields but where only a few are actually being matched or bound.
Most notably, matching a fixed slice would use a number of stack frames that
grows with the number of elements in the slice.
Fixes#17877.
I previously avoided `#[inline]`ing anything assuming someone would come in and explain to me where this would be appropriate. Apparently no one *really* knows, so I'll just go the opposite way an inline everything assuming someone will come in and yell at me that such-and-such shouldn't be `#[inline]`.
==================
For posterity, iteration comparisons:
```
test btree::map::bench::iter_20 ... bench: 971 ns/iter (+/- 30)
test btree::map::bench::iter_1000 ... bench: 29445 ns/iter (+/- 480)
test btree::map::bench::iter_100000 ... bench: 2929035 ns/iter (+/- 21551)
test treemap::bench::iter_20 ... bench: 530 ns/iter (+/- 66)
test treemap::bench::iter_1000 ... bench: 26287 ns/iter (+/- 825)
test treemap::bench::iter_100000 ... bench: 7650084 ns/iter (+/- 356711)
test trie::bench_map::iter_20 ... bench: 646 ns/iter (+/- 265)
test trie::bench_map::iter_1000 ... bench: 43556 ns/iter (+/- 5014)
test trie::bench_map::iter_100000 ... bench: 12988002 ns/iter (+/- 139676)
```
As you can see `btree` "scales" much better than `treemap`. `triemap` scales quite poorly.
Note that *completely* different results are given if the elements are inserted in order from the range [0, size]. In particular, TrieMap *completely* dominates in the sorted case. This suggests adding benches for both might be worthwhile. However unsorted is *probably* the more "normal" case, so I consider this "good enough" for now.
compiletest needs to link to native crate, or at least the `rt` library.
(I tried using a dependency on `rustrt` instead, and that did not resolve the problem. But this does.)
Partially addresses #17883
Only one warning remain, and I can't find a way to remove it without doing more bound checks:
```
shootout-nbody.rs:105:36: 105:51 warning: use of deprecated item: use iter_mut, #[warn(deprecated)] on by default
shootout-nbody.rs:105 let bi = match b_slice.mut_shift_ref() {
```
using `split_at_mut` may be an option, but it will do more bound checking.
If anyone have an idea, I'll update this PR.
Implement multidispatch and conditional dispatch. Because we do not attempt to preserve crate concatenation, this is a backwards compatible change. This is not yet fully integrated into method dispatch, so "UFCS"-style wrappers must be used to take advantage of the new features (see the run-pass tests).
cc #17307 (multidispatch)
cc #5527 (trait reform -- conditional dispatch)
Because we no longer preserve crate concatenability, this deviates slightly from what was specified in the RFC. The motivation for this change is described in [this blog post](http://smallcultfollowing.com/babysteps/blog/2014/09/30/multi-and-conditional-dispatch-in-traits/). I will post an amendment to the RFC in due course but do not anticipate great controversy on this point -- particularly as the RFCs more important features (e.g., conditional dispatch) just don't work without the change.
This change is an implementation of [RFC 69][rfc] which adds a third kind of
global to the language, `const`. This global is most similar to what the old
`static` was, and if you're unsure about what to use then you should use a
`const`.
The semantics of these three kinds of globals are:
* A `const` does not represent a memory location, but only a value. Constants
are translated as rvalues, which means that their values are directly inlined
at usage location (similar to a #define in C/C++). Constant values are, well,
constant, and can not be modified. Any "modification" is actually a
modification to a local value on the stack rather than the actual constant
itself.
Almost all values are allowed inside constants, whether they have interior
mutability or not. There are a few minor restrictions listed in the RFC, but
they should in general not come up too often.
* A `static` now always represents a memory location (unconditionally). Any
references to the same `static` are actually a reference to the same memory
location. Only values whose types ascribe to `Sync` are allowed in a `static`.
This restriction is in place because many threads may access a `static`
concurrently. Lifting this restriction (and allowing unsafe access) is a
future extension not implemented at this time.
* A `static mut` continues to always represent a memory location. All references
to a `static mut` continue to be `unsafe`.
This is a large breaking change, and many programs will need to be updated
accordingly. A summary of the breaking changes is:
* Statics may no longer be used in patterns. Statics now always represent a
memory location, which can sometimes be modified. To fix code, repurpose the
matched-on-`static` to a `const`.
static FOO: uint = 4;
match n {
FOO => { /* ... */ }
_ => { /* ... */ }
}
change this code to:
const FOO: uint = 4;
match n {
FOO => { /* ... */ }
_ => { /* ... */ }
}
* Statics may no longer refer to other statics by value. Due to statics being
able to change at runtime, allowing them to reference one another could
possibly lead to confusing semantics. If you are in this situation, use a
constant initializer instead. Note, however, that statics may reference other
statics by address, however.
* Statics may no longer be used in constant expressions, such as array lengths.
This is due to the same restrictions as listed above. Use a `const` instead.
[breaking-change]
Closes#17718
[rfc]: https://github.com/rust-lang/rfcs/pull/246
Updates the other_op function shared by the union/intersect/difference/symmetric_difference -with functions to fix an issue where certain elements would not be present in the result. To fix this, when other op is called, we resize self's nbits to account for any new elements that may be added to the set.
Example:
```rust
let mut a = BitvSet::new();
let mut b = BitvSet::new();
a.insert(0);
b.insert(5);
a.union_with(&b);
println!("{}", a); //Prints "{0}" instead of "{0, 5}"
```
parameter list.
This breaks code like:
fn f(a: int, a: int) { ... }
fn g<T,T>(a: T) { ... }
Change this code to not use the same name for a parameter. For example:
fn f(a: int, b: int) { ... }
fn g<T,U>(a: T) { ... }
Code like this is *not* affected, since `_` is not an identifier:
fn f(_: int, _: int) { ... } // OK
Closes#17568.
r? @alexcrichton
[breaking-change]
Instead of returning &'static [u8], an invocation of `bytes!()` now returns
`&'static [u8, ..N]` where `N` is the length of the byte vector. This should
functionally be the same, but there are some cases where an explicit cast may be
needed, so this is a:
[breaking-change]
