Avoid allocating extra copies of strings by using "" instead of ~"" for
the debug options list and for the `time` function. This is a small
change, but it is in a path that's always executed.
This purges about 500 lines of visitor cruft from lint passes. All lints are
handled in a much more sane way at this point. The other huge bonus of this
commit is that there are no more @-boxes in the lint passes, fixing the 500MB
memory regression seen when the lint passes were refactored.
Closes#8589
This purges about 500 lines of visitor cruft from lint passes. All lints are
handled in a much more sane way at this point. The other huge bonus of this
commit is that there are no more @-boxes in the lint passes, fixing the 500MB
memory regression seen when the lint passes were refactored.
Closes#8589
This makes sure that the top-level crate name is correct when emitting log
statements for a monomorphized function in another crate. This happens by
tracing the monomorphized ID back to the external source and then using that
crate index to get the name of the crate.
Closes#3046
This fixes two existing bugs along the way:
* The `transmute` intrinsic did not correctly handle casts of immediate
aggregates like newtype structs and tuples.
* The code for calling foreign functions used the wrong type to create
an `alloca` temporary
enum Foo { A, B }
fn foo() -> Foo { A }
Before:
; Function Attrs: nounwind uwtable
define void @_ZN3foo18hbedc642d5d9cf5aag4v0.0E(%enum.Foo* noalias nocapture sret, { i64, %tydesc*, i8*, i8*, i8 }* nocapture readnone) #0 {
"function top level":
%2 = getelementptr inbounds %enum.Foo* %0, i64 0, i32 0
store i64 0, i64* %2, align 8
ret void
}
After:
; Function Attrs: nounwind readnone uwtable
define %enum.Foo @_ZN3foo18hbedc642d5d9cf5aag4v0.0E({ i64, %tydesc*, i8*, i8*, i8 }* nocapture readnone) #0 {
"function top level":
ret %enum.Foo zeroinitializer
}
It is simply defined as `f64` across every platform right now.
A use case hasn't been presented for a `float` type defined as the
highest precision floating point type implemented in hardware on the
platform. Performance-wise, using the smallest precision correct for the
use case greatly saves on cache space and allows for fitting more
numbers into SSE/AVX registers.
If there was a use case, this could be implemented as simply a type
alias or a struct thanks to `#[cfg(...)]`.
Closes#6592
The mailing list thread, for reference:
https://mail.mozilla.org/pipermail/rust-dev/2013-July/004632.html
This is mostly an incremental change, picking off some uses of
@- or @mut-pointers that can be replaced by references.
Almost all of the builder functions in trans::build are updated,
mostly using `&Block` arguments instead of `@mut Block`.
Use &mut Block and &Block references where possible in the builder
functions in trans::build.
@mut Block remains in a few functions where I could not (not yet at
least) track down the runtime borrowck failures.
std::vec: Sane implementations for connect_vec and concat_vec
Avoid unnecessary copying of subvectors, and calculate the needed space
beforehand. These implementations are simple but better than the
previous.
Also only implement it once, for all `Vector<T>` using:
impl<'self, T: Clone, V: Vector<T>> VectorVector<T> for &'self [V]
Closes#9581
This is broken, and results in poor performance due to the undefined
behaviour in the LLVM IR. LLVM's `mergefunc` is a *much* better way of
doing this since it merges based on the equality of the bytecode.
For example, consider `std::repr`. It generates different code per
type, but is not included in the type bounds of generics.
The `mergefunc` pass works for most of our code but currently hits an
assert on libstd. It is receiving attention upstream so it will be
ready soon, but I don't think removing this broken code should wait any
longer. I've opened #9536 about enabling it by default.
Closes#8651Closes#3547Closes#2537Closes#6971Closes#9222
This is broken, and results in poor performance due to the undefined
behaviour in the LLVM IR. LLVM's `mergefunc` is a *much* better way of
doing this since it merges based on the equality of the bytecode.
For example, consider `std::repr`. It generates different code per
type, but is not included in the type bounds of generics.
The `mergefunc` pass works for most of our code but currently hits an
assert on libstd. It is receiving attention upstream so it will be
ready soon, but I don't think removing this broken code should wait any
longer. I've opened #9536 about enabling it by default.
Closes#8651Closes#3547Closes#2537Closes#6971Closes#9222
If an item is skipped due to it being unreachable or for some optimization, then
it shouldn't be encoded into the metadata (because it wasn't present in the
first place).
I have tried this fix and it seems to work either with single or multiple trait inheritance.
trait Base:Base2 + Base3{
fn foo(&self);
}
trait Base2 {
fn baz(&self);
}
trait Base3{
fn root(&self);
}
trait Super: Base{
fn bar(&self);
}
struct X;
impl Base for X {
fn foo(&self) {
println("base foo");
}
}
impl Base2 for X {
fn baz(&self) {
println("base2 baz");
}
}
impl Base3 for X {
fn root(&self) {
println("base3 root");
}
}
impl Super for X {
fn bar(&self) {
println("super bar");
}
}
fn main() {
let n = X;
let s = &n as &Super;
s.bar();
s.foo(); // super bar
s.baz();
s.root();
}
bmaxa@maxa:~/examples/rust$ rustc error.rs
bmaxa@maxa:~/examples/rust$ ./error
super bar
base foo
base2 baz
base3 root
This solves problem of incorrect indexing into vtable
when method from super trait was called through pointer
to derived trait.
Problem was that offset of super trait vtables
was not calculated at all.
Now it works, correct offset is calculated by
traversing all super traits up to super trait
where method belongs. That is how it is
intended to work.
r? anyone
Part of #7081.
Removed many unnecessary context arguments, turning them into visitors. Removed some @allocation.
If this lands, then I think the only thing left that is unaddressed are:
* the various lint visitors, and
* middle/privacy.rs, which has `impl<'self> Visitor<&'self method_map> for PrivacyVisitor`
This fixes private statics and functions from being usable cross-crates, along
with some bad privacy error messages. This is a reopening of #8365 with all the
privacy checks in privacy.rs instead of resolve.rs (where they should be
anyway).
These maps of exported items will hopefully get used for generating
documentation by rustdoc
Closes#8592
This fixes private statics and functions from being usable cross-crates, along
with some bad privacy error messages. This is a reopening of #8365 with all the
privacy checks in privacy.rs instead of resolve.rs (where they should be
anyway).
These maps of exported items will hopefully get used for generating
documentation by rustdoc
Closes#8592
r? anyone
Also got rid of a bit of `@mut` allocation. (Though not the monster that is `@mut FnCtxt`; that case is documented already on #7081; if we attack it, it will probably be its own ticket, not part of #7081.)
r? anyone.
Part of #7081.
More refactorings of the syntax::visit::Visitor implementations, folding so-called "environments" into the visitor impl when the latter was previously a trivial unit struct.
As usual, this refactoring only applies when the environments are not actually carrying state that is meant to be pushed and popped as we traverse the expression. (For an example where the environment *isn't* just passed through, see the `visit_fn` in `liveness.rs`.)
Got rid of a bit of @-allocation in borrowck.
Both cases should be pure-refactorings.
Resolves third bullet of #4691: if the functional-struct-update (FSU) expression `{ a: b, ..s }` causes `s` to move and `s` has a destructor, then the expression is illegal.
r? @nikomatsakis
This is for consistency in naming conventions.
- ``std::num::Float::NaN()`` is changed to ``nan()``;
- ``std::num::Float.is_NaN()`` is changed to ``is_nan()``; and
- ``std::num::strconv::NumStrConv::NaN()`` is changed to ``nan()``.
Fixes#9319.
This is the second of two parts of #8991, now possible as a new snapshot
has been made. (The first part implemented the unreachable!() macro; it
was #8992, 6b7b8f2682.)
``std::util::unreachable()`` is removed summarily; any code which used
it should now use the ``unreachable!()`` macro.
Closes#9312.
Closes#8991.
This is for consistency in naming conventions.
- ``std::num::Float::NaN()`` is changed to ``nan()``;
- ``std::num::Float.is_NaN()`` is changed to ``is_nan()``; and
- ``std::num::strconv::NumStrConv::NaN()`` is changed to ``nan()``.
Fixes#9319.
This is the second of two parts of #8991, now possible as a new snapshot
has been made. (The first part implemented the unreachable!() macro; it
was #8992, 6b7b8f2682.)
``std::util::unreachable()`` is removed summarily; any code which used
it should now use the ``unreachable!()`` macro.
Closes#9312.
Closes#8991.
std: Remove {float,f64,f32}::from_str in favor of from_str in the prelude
Like issue #9209, remove float::{from_str, from_str_radix} in favor of
the two corresponding traits. The same for modules f64 and f32.
New usage is:
from_str::<float>("1.2e34")
Like issue #9209, remove float::{from_str, from_str_radix} in favor of
the two corresponding traits. The same for modules f64 and f32.
New usage is
from_str::<float>("1.2e34")
If a static is flagged as address_insignificant, then for LLVM to actually
perform the relevant optimization it must have an internal linkage type. What
this means, though, is that the static will not be available to other crates.
Hence, if you have a generic function with an inner static, it will fail to link
when built as a library because other crates will attempt to use the inner
static externally.
This gets around the issue by inlining the static into the metadata. The same
relevant optimization is then applied separately in the external crate. What
this ends up meaning is that all statics tagged with #[address_insignificant]
will appear at most once per crate (by value), but they could appear in multiple
crates.
This should be the last blocker for using format! ...
This doesn't close any bugs as the goal is to convert the parameter to by-value, but this is a step towards being able to make guarantees about `&T` pointers (where T is Freeze) to LLVM.
In #8185 cross-crate condition handlers were fixed by ensuring that globals
didn't start appearing in different crates with different addressed. An
unfortunate side effect of that pull request is that constants weren't inlined
across crates (uint::bits is unknown to everything but libstd).
This commit fixes this inlining by using the `available_eternally` linkage
provided by LLVM. It partially reverts #8185, and then adds support for this
linkage type. The main caveat is that not all statics could be inlined into
other crates. Before this patch, all statics were considered "inlineable items",
but an unfortunate side effect of how we deal with `&static` and `&[static]`
means that these two cases cannot be inlined across crates. The translation of
constants was modified to propogate this condition of whether a constant
should be considered inlineable into other crates.
Closes#9036
also removes the unused `FastInvoke` wrapper, as it's never actually
going to be used (we can't *partially* switch to `fastcc`, and this is
only used for Rust functions)
The `noalias` attributes were being set only on function definitions,
not on all declarations. This is harmless for `noalias`, but prevented
some optimization opportunities and is *not* harmless for other
attributes like `sret` with ABI implications.
Closes#9104
In #8185 cross-crate condition handlers were fixed by ensuring that globals
didn't start appearing in different crates with different addressed. An
unfortunate side effect of that pull request is that constants weren't inlined
across crates (uint::bits is unknown to everything but libstd).
This commit fixes this inlining by using the `available_eternally` linkage
provided by LLVM. It partially reverts #8185, and then adds support for this
linkage type. The main caveat is that not all statics could be inlined into
other crates. Before this patch, all statics were considered "inlineable items",
but an unfortunate side effect of how we deal with `&static` and `&[static]`
means that these two cases cannot be inlined across crates. The translation of
constants was modified to propogate this condition of whether a constant
should be considered inlineable into other crates.
Closes#9036
While they may have the same name within various scopes, this changes static
names to use path_pretty_name to append some hash information at the end of the
symbol. We're then guaranteed that each static has a unique NodeId, so this
NodeId is as the "hash" of the pretty name.
Closes#9188
Beforehand it was assumed that the standard cdecl abi was used for all extern
fns of extern crates, but this reads the abi of the extern fn type and declares
the function in the local crate with the appropriate type.
I was trying to think of how to write a test for this, but I was just drawing up blanks :(. Are there standard functions in libc which are not of the cdecl abi? If so we could try linking to them and make sure that the cal completes successfully.
Otherwise, I manually verified that the function was declared correctly by looking at the llvm assembly.
cc #9055 (I'm not sure if this will fix that issue)
Beforehand it was assumed that the standard cdecl abi was used for all extern
fns of extern crates, but this reads the abi of the extern fn type and declares
the function in the local crate with the appropriate type.
Also fixed nasty bug caused by calling LLVMDIBuilderCreateStructType() with a null pointer where an empty array was expected (which would trigger an unintelligable assertion somewhere down the line).
While they may have the same name within various scopes, this changes static
names to use path_pretty_name to append some hash information at the end of the
symbol. We're then guaranteed that each static has a unique NodeId, so this
NodeId is as the "hash" of the pretty name.
Closes#9188
This is a series of patches to modernize option and result. The highlights are:
* rename `.unwrap_or_default(value)` and etc to `.unwrap_or(value)`
* add `.unwrap_or_default()` that uses the `Default` trait
* add `Default` implementations for vecs, HashMap, Option
* add `Option.and(T) -> Option<T>`, `Option.and_then(&fn() -> Option<T>) -> Option<T>`, `Option.or(T) -> Option<T>`, and `Option.or_else(&fn() -> Option<T>) -> Option<T>`
* add `option::ToOption`, `option::IntoOption`, `option::AsOption`, `result::ToResult`, `result::IntoResult`, `result::AsResult`, `either::ToEither`, and `either::IntoEither`, `either::AsEither`
* renamed `Option::chain*` and `Result::chain*` to `and_then` and `or_else` to avoid the eventual collision with `Iterator.chain`.
* Added a bunch of impls of `Default`
* Added a `#[deriving(Default)]` syntax extension
* Removed impls of `Zero` for `Option<T>` and vecs.
Since function pointers do not carry along the function attributes with
them in the type, this needs to be set on the call instruction itself.
Closes#9152
This is mostly for consistency, as you can now compare raw pointers in
constant expressions or without the standard library.
It also reduces the number of `ptrtoint` instructions in the IR, making
tracking down culprits of what's usually an anti-pattern easier.
This is mostly for consistency, as you can now compare raw pointers in
constant expressions or without the standard library.
It also reduces the number of `ptrtoint` instructions in the IR, making
tracking down culprits of what's usually an anti-pattern easier.
Who would have thought that namespaces are such a can of worms `:P` This is mostly because of some GDB idiosyncrasies (does not use namespace information but linkage-name attributes for displaying items contained in namespaces, also cannot handle functions lexically nested within functions), monomorphization, and information about external items only available from metadata.
This pull request tries to tackle the problem anyway:
* The `DW_AT_linkage_name` for functions is generated just to make GDB display a proper namespace-enabled function name. To this end, a pseudo-mangled name is generated, not corresponding to the real linkage name. This approach shows some success and could be extended to make GDB also show proper parameter types.
* As GDB won't accept subprogram DIEs nested within other subprogram DIEs, the `debuginfo` module now generates a *companion namespace* for each functions (iff needed). A function `fn abc()` will get a companion namespace with name `abc()`, which contains all items (modules, types, functions) declared within the functions scope. The real, proper solution, in my opinion, would be to faithfully reflect the program's lexical structure within DWARF (which allows arbitrary nesting of DIEs, afaik), but I am not sure LLVM's source level debugging implementation would like that and I am pretty sure GDB won't support this in the foreseeable future.
* Monomorphization leads to functions and companion namespaces like `somelib::some_func<int, float>()::some_other_function<bool, bool, bool>()`, which I think is the desired behaviour. There is some design space here, however. Maybe you people prefer `somelib::some_func()::some_other_function<bool, bool, bool>()` or `somelib::some_func()::some_other_function::<int, float, bool, bool, bool>()`.
The solution will work for now but there are a few things on my 'far future wish list':
* A real specification somewhere, what language constructs are mapped to what DWARF structures.
* Proper tests that directly compare the generated DWARF information to the expected results (possibly using something like [pyelftools](https://github.com/eliben/pyelftools) or llvm-dwarfdump)
* A unified implementation for crate-local and crate-external items (which would possibly involve beefing up `ast_map::path` and metadata a bit)
Any comments are welcome!
Closes#1541Closes#1542 (there might be other issues with function name prettiness, but this specific issue should be fixed)
Closes#7715 (source locations for structs and enums are now read correctly from the AST)
This way syntax extensions can generate unsafe blocks without worrying about
them generating unnecessary unsafe warnings. Perhaps a special keyword could be
added to be used in macros, but I don't think that's the best solution.
r? anyone
Remove some trivial Visitor structs, using their non-trivial Contexts as the Visitor implementation instead.
Removed a little bit of `@boxing` as well.
Part of ongoing work on #7081.
Ensures that each AST node has a unique id. Fixes numerous bugs in macro expansion and deriving. Add two
representative tests.
Fixes#7971Fixes#6304Fixes#8367Fixes#8754Fixes#8852Fixes#2543Fixes#7654
has a unique id. Fixes numerous bugs in macro expansion and deriving. Add two
representative tests.
Fixes#7971Fixes#6304Fixes#8367Fixes#8754Fixes#8852Fixes#2543Fixes#7654
Visit the free functions of std::vec and reimplement or remove some. Most prominently, remove `each_permutation` and replace with two iterators, ElementSwaps and Permutations.
Replace unzip, unzip_slice with an updated `unzip` that works with an iterator argument.
Replace each_permutation with a Permutation iterator. The new permutation iterator is more efficient since it uses an algorithm that produces permutations in an order where each is only one element swap apart, including swapping back to the original state with one swap at the end.
Unify the seldomly used functions `build`, `build_sized`, `build_sized_opt` into just one function `build`.
Remove `equal_sizes`
These functions have very few users since they are mostly replaced by
iterator-based constructions.
Convert a few remaining users in-tree, and reduce the number of
functions by basically renaming build_sized_opt to build, and removing
the other two. This for both the vec and the at_vec versions.
The basic construct x.len() == y.len() is just as simple.
This function used to be a precondition (not sure about the
terminology), so it had to be a function. This is not relevant any more.
Also redefine all of the standard logging macros to use more rust code instead
of custom LLVM translation code. This makes them a bit easier to understand, but
also more flexibile for future types of logging.
Additionally, this commit removes the LogType language item in preparation for
changing how logging is performed.
The trait will keep the `Iterator` naming, but a more concise module
name makes using the free functions less verbose. The module will define
iterables in addition to iterators, as it deals with iteration in
general.
This removes another large chunk of this odd 'clownshoes' identifier showing up
in symbol names. These all originated from external crates because the encoded
items were encoded independently of the paths calculated in ast_map. The
encoding of these paths now uses the helper function in ast_map to calculate the
"pretty name" for an impl block.
Unfortunately there is still no information about generics in the symbol name,
but it's certainly vastly better than before
hash::__extensions__::write::_version::v0.8
becomes
hash::Writer$SipState::write::hversion::v0.8
This also fixes bugs in which lots of methods would show up as `meth_XXX`, they
now only show up as `meth` and throw some extra characters onto the version
string.
This removes another large chunk of this odd 'clownshoes' identifier showing up
in symbol names. These all originated from external crates because the encoded
items were encoded independently of the paths calculated in ast_map. The
encoding of these paths now uses the helper function in ast_map to calculate the
"pretty name" for an impl block.
Unfortunately there is still no information about generics in the symbol name,
but it's certainly vastly better than before
hash::__extensions__::write::_version::v0.8
becomes
hash::Writer$SipState::write::hversion::v0.8
This also fixes bugs in which lots of methods would show up as `meth_XXX`, they
now only show up as `meth` and throw some extra characters onto the version
string.
Also redefine all of the standard logging macros to use more rust code instead
of custom LLVM translation code. This makes them a bit easier to understand, but
also more flexibile for future types of logging.
Additionally, this commit removes the LogType language item in preparation for
changing how logging is performed.
These commits fix bugs related to identically named statics in functions of implementations in various situations. The commit messages have most of the information about what bugs are being fixed and why.
As a bonus, while I was messing around with name mangling, I improved the backtraces we'll get in gdb by removing `__extensions__` for the trait/type being implemented and by adding the method name as well. Yay!
Remove __extensions__ from method symbols as well as the meth_XXX. The XXX is
now used to append a few characters at the end of the name of the symbol.
Closes#6602
This is currently unsound since `bool` is represented as `i8`. It will
become sound when `bool` is stored as `i8` but always used as `i1`.
However, the current behaviour will always be identical to `x & 1 != 0`,
so there's no need for it. It's also surprising, since `x != 0` is the
expected behaviour.
Closes#7311
gather_loans does not need to recurse into any items declared in the
current block. Rather than special-case `fk_item_fn` and `fk_method`,
just make the GatherLoanVisitor's visit_item method a no-op.
This indirectly implies that the example of #7740 is fixed:
fn f() {
static A: &'static char = &'A';
}
Since we do not recurse into items, we no longer encounter `&'A'`.
Significant progress on #6875, enough that I'll open new bugs and turn that into a metabug when this lands.
Description & example in the commit message.
Storing the type name in the `tydesc` aims to avoid the need to pass a type name in almost every single visitor method.
It would likely be much saner for `repr` to simply be passed the `TyDesc` corresponding to the function or just the type name, but this is good enough for now.
There are 6 new compiler recognised attributes: deprecated, experimental,
unstable, stable, frozen, locked (these levels are taken directly from
Node's "stability index"[1]). These indicate the stability of the
item to which they are attached; e.g. `#[deprecated] fn foo() { .. }`
says that `foo` is deprecated.
This comes with 3 lints for the first 3 levels (with matching names) that
will detect the use of items marked with them (the `unstable` lint
includes items with no stability attribute). The attributes can be given
a short text note that will be displayed by the lint. An example:
#[warn(unstable)]; // `allow` by default
#[deprecated="use `bar`"]
fn foo() { }
#[stable]
fn bar() { }
fn baz() { }
fn main() {
foo(); // "warning: use of deprecated item: use `bar`"
bar(); // all fine
baz(); // "warning: use of unmarked item"
}
The lints currently only check the "edges" of the AST: i.e. functions,
methods[2], structs and enum variants. Any stability attributes on modules,
enums, traits and impls are not checked.
[1]: http://nodejs.org/api/documentation.html
[2]: the method check is currently incorrect and doesn't work.
As with the previous commit, this is targeted at removing the possibility of
collisions between statics. The main use case here is when there's a
type-parametric function with an inner static that's compiled as a library.
Before this commit, any impl would generate a path item of "__extensions__".
This changes this identifier to be a "pretty name", which is either the last
element of the path of the trait implemented or the last element of the type's
path that's being implemented. That doesn't quite cut it though, so the (trait,
type) pair is hashed and again used to append information to the symbol.
Essentially, __extensions__ was removed for something nicer for debugging, and
then some more information was added to symbol name by including a hash of the
trait being implemented and type it's being implemented for. This should prevent
colliding names for inner statics in regular functions with similar names.
Whenever a generic function was encountered, only the top-level items were
recursed upon, even though the function could contain items inside blocks or
nested inside of other expressions. This fixes the existing code from traversing
just the top level items to using a Visitor to deeply recurse and find any items
which need to be translated.
This was uncovered when building code with --lib, because the encode_symbol
function would panic once it found that an item hadn't been translated.
Closes#8134
Whenever a generic function was encountered, only the top-level items were
recursed upon, even though the function could contain items inside blocks or
nested inside of other expressions. This fixes the existing code from traversing
just the top level items to using a Visitor to deeply recurse and find any items
which need to be translated.
This was uncovered when building code with --lib, because the encode_symbol
function would panic once it found that an item hadn't been translated.
Closes#8134
This removes the stacking of type parameters that occurs when invoking
trait methods, and fixes all places in the standard library that were
relying on it. It is somewhat awkward in places; I think we'll probably
want something like the `Foo::<for T>::new()` syntax.
Fixes for #8625 to prevent assigning to `&mut` in borrowed or aliasable locations. The old code was insufficient in that it failed to catch bizarre cases like `& &mut &mut`.
r? @pnkfelix
Beforehand, it was unclear whether rust was performing the "recommended set" of
optimizations provided by LLVM for code. This commit changes the way we run
passes to closely mirror that of clang, which in theory does it correctly. The
notable changes include:
* Passes are no longer explicitly added one by one. This would be difficult to
keep up with as LLVM changes and we don't guaranteed always know the best
order in which to run passes
* Passes are now managed by LLVM's PassManagerBuilder object. This is then used
to populate the various pass managers run.
* We now run both a FunctionPassManager and a module-wide PassManager. This is
what clang does, and I presume that we *may* see a speed boost from the
module-wide passes just having to do less work. I have no measured this.
* The codegen pass manager has been extracted to its own separate pass manager
to not get mixed up with the other passes
* All pass managers now include passes for target-specific data layout and
analysis passes
Some new features include:
* You can now print all passes being run with `-Z print-llvm-passes`
* When specifying passes via `--passes`, the passes are now appended to the
default list of passes instead of overwriting them.
* The output of `--passes list` is now generated by LLVM instead of maintaining
a list of passes ourselves
* Loop vectorization is turned on by default as an optimization pass and can be
disabled with `-Z no-vectorize-loops`
All of these "copies" of clang are based off their [source code](http://clang.llvm.org/doxygen/BackendUtil_8cpp_source.html) in case anyone is curious what my source is. I was hoping that this would fix#8665, but this does not help the performance issues found there. Hopefully i'll allow us to tweak passes or see what's going on to try to debug that problem.
Beforehand, it was unclear whether rust was performing the "recommended set" of
optimizations provided by LLVM for code. This commit changes the way we run
passes to closely mirror that of clang, which in theory does it correctly. The
notable changes include:
* Passes are no longer explicitly added one by one. This would be difficult to
keep up with as LLVM changes and we don't guaranteed always know the best
order in which to run passes
* Passes are now managed by LLVM's PassManagerBuilder object. This is then used
to populate the various pass managers run.
* We now run both a FunctionPassManager and a module-wide PassManager. This is
what clang does, and I presume that we *may* see a speed boost from the
module-wide passes just having to do less work. I have no measured this.
* The codegen pass manager has been extracted to its own separate pass manager
to not get mixed up with the other passes
* All pass managers now include passes for target-specific data layout and
analysis passes
Some new features include:
* You can now print all passes being run with `-Z print-llvm-passes`
* When specifying passes via `--passes`, the passes are now appended to the
default list of passes instead of overwriting them.
* The output of `--passes list` is now generated by LLVM instead of maintaining
a list of passes ourselves
* Loop vectorization is turned on by default as an optimization pass and can be
disabled with `-Z no-vectorize-loops`
This patchset enables rustc to cross-build mingw-w64 outputs.
Tested on mingw + mingw-w64 (mingw-builds, win64/seh/win32-threads/gcc-4.8.1).
I also patched llvm to support Win64 stack unwinding.
ebe22bdbce
I cross-built test/run-pass/smallest-hello-world.rs and confirmed it works.
However, I also found something went wrong if I don't have custom `#[start]` routine.
Further followup on #7081.
There still remains writeback.rs, but I want to wait to investigate that one because I've seen `make check` issues with it in the past.
