When reuing a definition across codegen units, we obviously cannot use
internal linkage, but using external linkage means that we can end up
with multiple conflicting definitions of a single symbol across
multiple crates. Since the definitions should all be equal
semantically, we can use weak_odr linkage to resolve the situation.
Fixes#32518
We use a 64bit integer to pass the set of attributes that is to be
removed, but the called C function expects a 32bit integer. On most
platforms this doesn't cause any problems other than being unable to
unset some attributes, but on ARM even the lower 32bit aren't handled
correctly because the 64bit value is passed in different registers, so
the C function actually sees random garbage.
So we need to fix the relevant functions to use 32bit integers instead.
Additionally we need an implementation that actually accepts 64bit
integers because some attributes can only be unset that way.
Fixes#32360
`fast` a.k.a UnsafeAlgebra is the flag for enabling all "unsafe"
(according to llvm) float optimizations.
See LangRef for more information http://llvm.org/docs/LangRef.html#fast-math-flags
Providing these operations with less precise associativity rules (for
example) is useful to numerical applications.
For example, the summation loop:
let sum = 0.;
for element in data {
sum += *element;
}
Using the default floating point semantics, this loop expresses the
floats must be added in a sequence, one after another. This constraint
is usually completely unintended, and it means that no autovectorization
is possible.
Hopefully the author caught all the cases. For the mir_dynamic_drops_3 test case the ratio of
memsets to other instructions is 12%. On the other hand we actually do not double drop for at least
the test cases provided anymore in MIR.
Rust currently emits atomic loads and stores with the LLVM `volatile` qualifier. This is unnecessary and prevents LLVM from performing optimization on these atomic operations.
This commit transitions the compiler to using the new exception handling
instructions in LLVM for implementing unwinding for MSVC. This affects both 32
and 64-bit MSVC as they're both now using SEH-based strategies. In terms of
standard library support, lots more details about how SEH unwinding is
implemented can be found in the commits.
In terms of trans, this change necessitated a few modifications:
* Branches were added to detect when the old landingpad instruction is used or
the new cleanuppad instruction is used to `trans::cleanup`.
* The return value from `cleanuppad` is not stored in an `alloca` (because it
cannot be).
* Each block in trans now has an `Option<LandingPad>` instead of `is_lpad: bool`
for indicating whether it's in a landing pad or not. The new exception
handling intrinsics require that on MSVC each `call` inside of a landing pad
is annotated with which landing pad that it's in. This change to the basic
block means that whenever a `call` or `invoke` instruction is generated we
know whether to annotate it as part of a cleanuppad or not.
* Lots of modifications were made to the instruction builders to construct the
new instructions as well as pass the tagging information for the call/invoke
instructions.
* The translation of the `try` intrinsics for MSVC has been overhauled to use
the new `catchpad` instruction. The filter function is now also a
rustc-generated function instead of a purely libstd-defined function. The
libstd definition still exists, it just has a stable ABI across architectures
and leaves some of the really weird implementation details to the compiler
(e.g. the `localescape` and `localrecover` intrinsics).
This brings some routine upgrades to the bundled LLVM that we're using, the most
notable of which is a bug fix to the way we handle range asserts when loading
the discriminant of an enum. This fix ended up being very similar to f9d4149c
where we basically can't have a range assert when loading a discriminant due to
filling drop, and appropriate flags were added to communicate this to
`trans::adt`.
Travis CI has new infrastructure using the Google Compute Engine which has both
faster CPUs and more memory, and we've been encouraged to switch as it should
help our build times! The only downside currently, however, is that IPv6 is
disabled, causing a number of standard library tests to fail.
Consequently this commit tweaks our travis config in a few ways:
* ccache is disabled as it's not working on GCE just yet
* Docker is used to run tests inside which reportedly will get IPv6 working
* A system LLVM installation is used instead of building LLVM itself. This is
primarily done to reduce build times, but we want automation for this sort of
behavior anyway and we can extend this in the future with building from source
as well if needed.
* gcc-specific logic is removed as the docker image for Ubuntu gives us a
recent-enough gcc by default.
This commit moves the IR files in the distribution, rust_try.ll,
rust_try_msvc_64.ll, and rust_try_msvc_32.ll into the compiler from the main
distribution. There's a few reasons for this change:
* LLVM changes its IR syntax from time to time, so it's very difficult to
have these files build across many LLVM versions simultaneously. We'll likely
want to retain this ability for quite some time into the future.
* The implementation of these files is closely tied to the compiler and runtime
itself, so it makes sense to fold it into a location which can do more
platform-specific checks for various implementation details (such as MSVC 32
vs 64-bit).
* This removes LLVM as a build-time dependency of the standard library. This may
end up becoming very useful if we move towards building the standard library
with Cargo.
In the immediate future, however, this commit should restore compatibility with
LLVM 3.5 and 3.6.
The C API of this function changed so it no longer takes a personality function.
A shim was introduced to call the right LLVM function (depending on which
version we're compiled against) to set the personality function on the outer
function.
The compiler only ever sets one personality function for all generated
functions, so this should be equivalent.
We have previously always relied upon an external tool, `ar`, to modify archives
that the compiler produces (staticlibs, rlibs, etc). This approach, however, has
a number of downsides:
* Spawning a process is relatively expensive for small compilations
* Encoding arguments across process boundaries often incurs unnecessary overhead
or lossiness. For example `ar` has a tough time dealing with files that have
the same name in archives, and the compiler copies many files around to ensure
they can be passed to `ar` in a reasonable fashion.
* Most `ar` programs found do **not** have the ability to target arbitrary
platforms, so this is an extra tool which needs to be found/specified when
cross compiling.
The LLVM project has had a tool called `llvm-ar` for quite some time now, but it
wasn't available in the standard LLVM libraries (it was just a standalone
program). Recently, however, in LLVM 3.7, this functionality has been moved to a
library and is now accessible by consumers of LLVM via the `writeArchive`
function.
This commit migrates our archive bindings to no longer invoke `ar` by default
but instead make a library call to LLVM to do various operations. This solves
all of the downsides listed above:
* Archive management is now much faster, for example creating a "hello world"
staticlib is now 6x faster (50ms => 8ms). Linking dynamic libraries also
recently started requiring modification of rlibs, and linking a hello world
dynamic library is now 2x faster.
* The compiler is now one step closer to "hassle free" cross compilation because
no external tool is needed for managing archives, LLVM does the right thing!
This commit does not remove support for calling a system `ar` utility currently.
We will continue to maintain compatibility with LLVM 3.5 and 3.6 looking forward
(so the system LLVM can be used wherever possible), and in these cases we must
shell out to a system utility. All nightly builds of Rust, however, will stop
needing a system `ar`.
This commit updates the LLVM submodule in use to the current HEAD of the LLVM
repository. This is primarily being done to start picking up unwinding support
for MSVC, which is currently unimplemented in the revision of LLVM we are using.
Along the way a few changes had to be made:
* As usual, lots of C++ debuginfo bindings in LLVM changed, so there were some
significant changes to our RustWrapper.cpp
* As usual, some pass management changed in LLVM, so clang was re-scrutinized to
ensure that we're doing the same thing as clang.
* Some optimization options are now passed directly into the
`PassManagerBuilder` instead of through CLI switches to LLVM.
* The `NoFramePointerElim` option was removed from LLVM, favoring instead the
`no-frame-pointer-elim` function attribute instead.
Additionally, LLVM has picked up some new optimizations which required fixing an
existing soundness hole in the IR we generate. It appears that the current LLVM
we use does not expose this hole. When an enum is moved, the previous slot in
memory is overwritten with a bit pattern corresponding to "dropped". When the
drop glue for this slot is run, however, the switch on the discriminant can
often start executing the `unreachable` block of the switch due to the
discriminant now being outside the normal range. This was patched over locally
for now by having the `unreachable` block just change to a `ret void`.
This commit introduce a third parameter for compatible_ifn!, as new
intrinsics are being added in recent LLVM releases and there is no
need to hardcode a specific case.
Signed-off-by: Luca Bruno <lucab@debian.org>
For imports of constants across DLLs to work on Windows it *requires* that the
import be marked with `dllimport` (unlike functions where the marker is
optional, but strongly recommended). This currently isn't working for importing
FFI constants across boundaries, however, so the one constant exported from
`rustc_llvm.dll` is now a function to be called instead.
These new intrinsics are comparable to `atomic_signal_fence` in C++,
ensuring the compiler will not reorder memory accesses across the
barrier, nor will it emit any machine instructions for it.
Closes#24118, implementing RFC 888.
When linking an archive statically to an rlib, the compiler will extract all
contents of the archive and add them all to the rlib being generated. The
current method of extraction is to run `ar x`, dumping all files into a
temporary directory. Object archives, however, are allowed to have multiple
entries with the same file name, so there is no method for them to extract their
contents into a directory in a lossless fashion.
This commit adds iterator support to the `ArchiveRO` structure which hooks into
LLVM's support for reading object archives. This iterator is then used to
inspect each object in turn and extract it to a unique location for later
assembly.
LLVM older that 3.6 has a bug that cause assertions when compiling certain
constructs. For 3.5 there's still a chance that the bug might get fixed
in 3.5.2, so let's keep allowing to compile with it for it for now.
Apparently the default getFile implementation for a memory buffer in LLVM ends
up requiring a null terminator at the end of the file. This isn't true a good
bit of the time apparently on OSX. There have been a number of failed
nightly/snapshot builds recently with this strange assertion.
This modifies the calls to MemoryBuffer::getFile to explicitly not ask for a
null terminator.
To fix#8106, we need an LLVM version that contains r211082 aka 0dee6756
which fixes a bug that blocks that issue.
There have been some tiny API changes in LLVM, and cmpxchg changed its
return type. The i1 part of the new return type is only interesting when
using the new weak cmpxchg, which we don't do.
LLVM internally uses `uint64_t` for array size, but the corresponding
C API (`LLVMArrayType`) uses `unsigned int` so ths value is truncated.
Therefore rustc generates wrong type for fixed-sized large vector e.g.
`[0 x i8]` for `[0u8, ..(1 << 32)]`.
This patch adds `LLVMRustArrayType` function for `uint64_t` support.
Many of the instances of setting a global error variable ended up leaving a
dangling pointer into free'd memory. This changes the method of error
transmission to strdup any error and "relinquish ownership" to rustc when it
gets an error. The corresponding Rust code will then free the error as
necessary.
Closes#12865
In upgrading LLVM, only rust functions had the "split-stack" attribute added.
This commit changes the addition of LLVM's "split-stack" attribute to *always*
occur and then we remove it sometimes if the "no_split_stack" rust attribute is
present.
Closes#13625
This comes with a number of fixes to be compatible with upstream LLVM:
* Previously all monomorphizations of "mem::size_of()" would receive the same
symbol. In the past LLVM would silently rename duplicated symbols, but it
appears to now be dropping the duplicate symbols and functions now. The symbol
names of monomorphized functions are now no longer solely based on the type of
the function, but rather the type and the unique hash for the
monomorphization.
* Split stacks are no longer a global feature controlled by a flag in LLVM.
Instead, they are opt-in on a per-function basis through a function attribute.
The rust #[no_split_stack] attribute will disable this, otherwise all
functions have #[split_stack] attached to them.
* The compare and swap instruction now takes two atomic orderings, one for the
successful case and one for the failure case. LLVM internally has an
implementation of calculating the appropriate failure ordering given a
particular success ordering (previously only a success ordering was
specified), and I copied that into the intrinsic translation so the failure
ordering isn't supplied on a source level for now.
* Minor tweaks to LLVM's API in terms of debuginfo, naming, c++11 conventions,
etc.
The recent pull request to remove libc from libstd has hit a wall in compiling
on windows, and I've been trying to investigate on the try bots as to why (it
compiles locally just fine). To the best of my knowledge, the LLVM section
iterator is behaving badly when iterating over the sections of the libc DLL.
Upon investigating the LLVMGetSectionName function in LLVM, I discovered that
this function doesn't always return a null-terminated string. It returns the
data pointer of a StringRef instance (LLVM's equivalent of &str essentially),
but it has no method of returning the length of the name of the section.
This commit modifies the section iteration when loading libraries to invoke a
custom LLVMRustGetSectionName which will correctly return both the length and
the data pointer.
I have not yet verified that this will fix landing liblibc, as it will require a
snapshot before doing a full test. Regardless, this is a worrisome situation
regarding the LLVM API, and should likely be fixed anyway.
The llvm.copysign and llvm.round intrinsics weren't added until LLVM 3.4, so if
we're on LLVM 3.3 we lower these to calls in libm instead of LLVM intrinsics.
This should fix our travis failures.
The travis builds have been breaking recently because LLVM 3.5 upstream is
changing. This looks like it's likely to continue, so it would be more useful
for us if we could lock ourselves to a system LLVM version that is not changing.
This commit has the support to bring our C++ glue to LLVM back in line with what
was possible back in LLVM 3.{3,4}. I don't think we're going to be able to
reasonably protect against regressions in the future, but this kind of code is a
good sign that we can continue to use the system LLVM for simple-ish things.
Codegen for ARM won't work and it won't have some of the perf improvements we
have, but using the system LLVM should work well enough for development.
Set "Dwarf Version" to 2 on OS X to avoid toolchain incompatibility, and
set "Debug Info Version" to prevent debug info from being stripped from
bitcode.
Fixes#11352.
Major changes:
- Define temporary scopes in a syntax-based way that basically defaults
to the innermost statement or conditional block, except for in
a `let` initializer, where we default to the innermost block. Rules
are documented in the code, but not in the manual (yet).
See new test run-pass/cleanup-value-scopes.rs for examples.
- Refactors Datum to better define cleanup roles.
- Refactor cleanup scopes to not be tied to basic blocks, permitting
us to have a very large number of scopes (one per AST node).
- Introduce nascent documentation in trans/doc.rs covering datums and
cleanup in a more comprehensive way.
This pull request fixes#11083. The problem was that recursive type definitions were not properly handled for enum types, leading to problems with LLVM's metadata "uniquing". This bug has already been fixed for struct types some time ago (#9658) but I seem to have forgotten about enums back then. I added the offending code from issue #11083 as a test case.
We were previously reading metadata via `ar p`, but as learned from rustdoc
awhile back, spawning a process to do something is pretty slow. Turns out LLVM
has an Archive class to read archives, but it cannot write archives.
This commits adds bindings to the read-only version of the LLVM archive class
(with a new type that only has a read() method), and then it uses this class
when reading the metadata out of rlibs. When you put this in tandem of not
compressing the metadata, reading the metadata is 4x faster than it used to be
The timings I got for reading metadata from the respective libraries was:
libstd-04ff901e-0.9-pre.dylib => 100ms
libstd-04ff901e-0.9-pre.rlib => 23ms
librustuv-7945354c-0.9-pre.dylib => 4ms
librustuv-7945354c-0.9-pre.rlib => 1ms
librustc-5b94a16f-0.9-pre.dylib => 87ms
librustc-5b94a16f-0.9-pre.rlib => 35ms
libextra-a6ebb16f-0.9-pre.dylib => 63ms
libextra-a6ebb16f-0.9-pre.rlib => 15ms
libsyntax-2e4c0458-0.9-pre.dylib => 86ms
libsyntax-2e4c0458-0.9-pre.rlib => 22ms
In order to always take advantage of these faster metadata read-times, I sort
the files in filesearch based on whether they have an rlib extension or not
(prefer all rlib files first).
Overall, this halved the compile time for a `fn main() {}` crate from 0.185s to
0.095s on my system (when preferring dynamic linking). Reading metadata is still
the slowest pass of the compiler at 0.035s, but it's getting pretty close to
linking at 0.021s! The next best optimization is to just not copy the metadata
from LLVM because that's the most expensive part of reading metadata right now.
This commit implements LTO for rust leveraging LLVM's passes. What this means
is:
* When compiling an rlib, in addition to insdering foo.o into the archive, also
insert foo.bc (the LLVM bytecode) of the optimized module.
* When the compiler detects the -Z lto option, it will attempt to perform LTO on
a staticlib or binary output. The compiler will emit an error if a dylib or
rlib output is being generated.
* The actual act of performing LTO is as follows:
1. Force all upstream libraries to have an rlib version available.
2. Load the bytecode of each upstream library from the rlib.
3. Link all this bytecode into the current LLVM module (just using llvm
apis)
4. Run an internalization pass which internalizes all symbols except those
found reachable for the local crate of compilation.
5. Run the LLVM LTO pass manager over this entire module
6a. If assembling an archive, then add all upstream rlibs into the output
archive. This ignores all of the object/bitcode/metadata files rust
generated and placed inside the rlibs.
6b. If linking a binary, create copies of all upstream rlibs, remove the
rust-generated object-file, and then link everything as usual.
As I have explained in #10741, this process is excruciatingly slow, so this is
*not* turned on by default, and it is also why I have decided to hide it behind
a -Z flag for now. The good news is that the binary sizes are about as small as
they can be as a result of LTO, so it's definitely working.
Closes#10741Closes#10740
LLVM's JIT has been updated numerous times, and we haven't been tracking it at
all. The existing LLVM glue code no longer compiles, and the JIT isn't used for
anything currently.
This also rebases out the FixedStackSegment support which we have added to LLVM.
None of this is still in use by the compiler, and there's no need to keep this
functionality around inside of LLVM.
This is needed to unblock #10708 (where we're tripping an LLVM assertion).
Example:
void ({ i64, %tydesc*, i8*, i8*, i8 }*, i64*, %"struct.std::fmt::Formatter[#1]"*)*
Before, we would print 20 levels deep due to recursion in the type
definition.
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).
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`
We currently have no need for the frame pointers on any platform. They
may eventually be needed on platforms without an equivalent to the DWARF
call frame information to walk the stack in the garbage collector.
Closes#7477
The first commit message is pretty good, but whomever reviews this should probably also at least glance at the changes I made in LLVM. I basically reorganized our pending patch queue to be a bit more organized and clearer in what needs to go where. After this, our queue would be:
* Add the `no-split-stack` attribute
* Add the `fixedstacksegment` attribute
* Add split-stacks for arm android
* Add split-stacks for arm linux
* Add split stacks for mips
Then there's a patch which I added to get rust to build at all on LLVM-head, and I'm not quite sure why it's there, but nothing seems to be crashing for now! (famous last words).
Otherwise, I just updated code to reflect the changes I made in LLVM with the only major change being the advent of the new `no_split_stack` attribute. This is work towards #1226, but someone more familiar with the code should probably actually assign the attribute to the appropriate functions.
Also as a bonus, I've verified that this closes#5774
Adds `--target-cpu` flag which lets you choose a more specific target cpu instead of just passing the default, `generic`. It's more or less akin to `-mcpu`/`-mtune` in clang/gcc.
This can be applied to statics and it will indicate that LLVM will attempt to
merge the constant in .data with other statics.
I have preliminarily applied this to all of the statics generated by the new
`ifmt!` syntax extension. I compiled a file with 1000 calls to `ifmt!` and a
separate file with 1000 calls to `fmt!` to compare the sizes, and the results
were:
fmt 310k
ifmt (before) 529k
ifmt (after) 202k
This now means that ifmt! is both faster and smaller than fmt!, yay!
* LLVM now has a C interface to LLVMBuildAtomicRMW
* The exception handling support for the JIT seems to have been dropped
* Various interfaces have been added or headers have changed
Refactor the optimization passes to explicitly use the passes. This commit
just re-implements the same passes as were already being run.
It also adds an option (behind `-Z`) to run the LLVM lint pass on the
unoptimized IR.
The default versions (atomic_load and atomic_store) are sequentially consistent.
The atomic_load_acq intrinsic acquires as described in [1].
The atomic_store_rel intrinsic releases as described in [1].
[1]: http://llvm.org/docs/Atomics.html
In my WIP on rustpkg, I was calling driver code that calls
LLVMRustWriteOutputFile more than once. This was making LLVM
unhappy, since that function has code that initializes the
command-line options for LLVM, and I guess you can't do that more
than once. So, check if they've already been initialized.
adjusting a few foreign functions that were declared with by-ref
mode. This also allows us to remove by-val mode in the near future.
With copy mode, though, we have to be careful because Rust will implicitly pass
somethings by pointer but this may not be the C ABI rules. For example, rust
will pass a struct Foo as a Foo*. So I added some code into the adapters to
fix this (though the C ABI rules may put the pointer back, oh well).
This patch also includes a lint mode for the use of by-ref mode
in foreign functions as the semantics of this have changed.
DebugFlag is conditionally exported by LLVM in llvm/Support/Debug.h
in-between an #ifndef NDEBUG block; RustWrapper should not
unconditionally use it. This closes#3701.
Signed-off-by: Luca Bruno <lucab@debian.org>
rustc generates output files in LLVM bitcode format if "--emit-llvm"
option is given. When used with the "-S" option, rustc generates LLVM
intermediate language assembly files.
Fixes Issue #476
